An attempt to reauthorize Section 702, the so-called crown jewel of US spy powers, failed for a third time in the House of Representatives after former president Donald Trump criticized the law.

For the third time since December, House Speaker Mike Johnson has failed to wrangle support for reauthorizing a critical US surveillance program, raising questions about the future of a law that compels certain businesses to wiretap foreigners on the government’s behalf.

Johnson lost 19 Republicans on Tuesday in a procedural vote that traditionally falls along party lines. Republicans control the House of Representatives but only by a razor-thin margin. The failed vote comes just hours after former US president Donald Trump ordered Republicans to “Kill FISA” in a 2 am post on Truth Social, referring to the Foreign Intelligence Surveillance Act, under which the program is authorized.

The Section 702 surveillance program, which targets foreigners overseas while sweeping up a large amount of US communications as well, is set to sunset on April 19. The program was extended by four months in late December following Johnson’s first failed attempt to hold a vote.

Congressional sources tell WIRED they have no idea what the next steps will be.

The program itself will carry on into the next year, regardless of whether Johnson manages to muster up another vote in the next week. Congress does not directly authorize the surveillance. Instead, it allows the US intelligence services to seek “certifications” from a secret surveillance court on a yearly basis.

The Justice Department applied for new certifications in February. Last week, it announced they’d been approved by the court. The government’s power to issue new directives under the program without Congress’s approval, however, remains in question.

The certifications, which are required only due to the “incidental” collection of US calls, generally permit the program’s use in cases involving terrorism, cybercrime, and weapons proliferation. US intelligence officials have also touted the program as crucial in combating the flood of fentanyl-related substances entering the US from overseas.

The program remains controversial due to a laundry list of abuses committed primarily at the Federal Bureau of Investigation, which maintains a database that holds a portion of the raw data collected under 702.

Although the government says it only “targets” foreigners, it has acknowledged collecting a large amount of US communications in the process. (The actual amount, it says, is impossible to calculate.) Nevertheless, it claims that once those communications are in the government’s possession, it is constitutional for federal agents to review those wiretaps without a warrant.

An unlikely coalition of progressives and conservative lawmakers formed last year in a push to end these warrantless searches, many of the Republicans involved vocal critics of the FBI following its misuse of FISA to target a Trump campaign staffer in 2016. (The 702 program, which is only one part of FISA, was not implicated in that particular controversy.)

Privacy experts have criticized proposed changes to the Section 702 program championed by members of the House Intelligence Committee, as well as Johnson, who had previously voted in favor of a warrant requirement despite now opposing it.

“It seems Congressional leadership needs to be reminded that these privacy protections are overwhelmingly popular,” says Sean Vitka, policy director at Demand Progress, a civil liberties–focused nonprofit. “Surveillance reformers remain willing and able to do that.”

A group of attorneys—among the few to ever present arguments before the Foreign Intelligence Surveillance Court—said in a statement on Tuesday that an amendment offered up by the Intel committee risked dramatically increasing the number of US businesses forced to cooperate with the program.

Declassified filings released by the FISA court last year revealed that the FBI had misused the 702 program more than 278,000 times, including, as reported by _The Washington Post_, against “crime victims, January 6 riot suspects, people arrested at protests after the policing killing of George Floyd in 2020 and—in one case—19,000 donors to a congressional candidate.”

James Czerniawaski, a senior policy analyst at Americans for Prosperity, a Washington, DC, think tank pushing for changes to Section 702, says that despite recognizing its value, it remained a “troubled program” in need of “significant and meaningful reforms.”

“The outcome of today was completely avoidable,” he says, “but it requires the Intelligence Community and its allies to recognize that its days of unaccountable and unconditional spying on Americans are over.”

Trump Loyalists Kill Vote on US Wiretap Program

It's finally happening: Rather than just for productivity and research, threat actors are using LLMs to write malware. But companies need not worry just yet.

Source: Ole.CNX via Shutterstock

Researchers from Proofpoint recently observed a malicious campaign targeting dozens of organizations across various industries in Germany. One part of the attack chain stood out in particular: an otherwise ordinary malware dropper whose code had clearly been generated by artificial intelligence (AI).

What the researchers discovered: Initial access broker (IAB) TA547 is using the AI-generated dropper in phishing attacks.

Though it may be a harbinger of more to come, it's no cause for panic. Defending against malware is the same no matter who or what writes it, and AI malware isn't likely to take over the world just yet.

"For the next few years, I don't see malware coming out of LLMs being more sophisticated than something a human is going to be able to write," says Daniel Blackford, senior manager of threat research at Proofpoint. After all, AI aside, "We've got very talented software engineers who are adversarially working against us."

TA547 has a long history of financially motivated cyberattacks. It came to prominence trafficking Trickbot, but has cycled through handfuls of other popular cybercrime tools, including Gozi/Ursnif, Lumma stealer, NetSupport RAT, StealC, ZLoader, and more.

"We're seeing — not just with TA547, but with other groups as well — much faster iteration through development cycles, adoption of other malware, trying new techniques to see what will stick," Blackford explains. And TA547's latest evolution seems to have been with AI.

Its attacks began with brief impersonation emails — for example, masquerading as the German retail company Metro AG. The emails contained password-protected ZIP files couching compressed LNK files. The LNK files, when executed, triggered a Powershell script that dropped the Rhadamanthys infostealer.

Sounds simple enough, but the Powershell script that dropped Rhadamanthys had one strange characteristic. Within the code, above each and every component, was a hashtag followed by hyper-specific comments about what the component achieved.

A snippet of the TA547 dropper. Source: Proofpoint

As Proofpoint noted, this is characteristic of LLM-generated code, indicating that the group — or whoever originally wrote the dropper — used some sort of chatbot to write it.

**Is Worse AI Malware to Come?**

Like the rest of us, cyberattackers have been experimenting with how AI chatbots can help them achieve their goals more easily, expeditiously, and effectively.

Some have figured out little ways to use AI to enhance their day-to-day operations, for example by aiding research into targets and emerging vulnerabilities. But aside from proofs-of-concept and the odd novelty tool, there hasn't been much evidence that hackers are writing useful malware with the help of AI.

That, Blackford says, is because humans are still far better than robots at writing malicious code. Plus, AI developers have taken steps to prevent the misuse of their software.

At least for now, he says, "the ways that these groups are going to leverage AI to scale up their operations is more of an interesting problem than the idea that they're going to create some new super malware with it."

And even once they do autogenerate super malware, the job of defending against it will remain the same. As Proofpoint concluded in its post, "In the same way LLM-generated phishing emails to conduct business email compromise (BEC) use the same characteristics of human-generated content and are caught by automated detections, malware or scripts that incorporate machine-generated code will still run the same way in a sandbox (or on a host), triggering the same automated defenses."

**About the Author(s)**

Nate Nelson is a freelance writer based in New York City. Formerly a reporter at Threatpost, he contributes to a number of cybersecurity blogs and podcasts. He writes "Malicious Life" -- an award-winning Top 20 tech podcast on Apple and Spotify -- and hosts every other episode, featuring interviews with leading voices in security. He also co-hosts "The Industrial Security Podcast," the most popular show in its field.

TA547 Uses an LLM-Generated Dropper to Infect German Orgs

There are also two out-of-bounds write vulnerabilities in the AMD Radeon user mode driver for DirectX 11.

Wednesday, April 10, 2024 12:56

Cisco Talos’ Vulnerability Research team has disclosed 10 vulnerabilities over the past three weeks, including four in a line of TP-Link routers, one of which could allow an attacker to reset the devices’ settings back to the factory default. 

A popular open-source software for internet-of-things (IoT) and industrial control systems (ICS) networks also contains multiple vulnerabilities that could be used to arbitrarily create new files on the affected systems or overwrite existing ones. 

For Snort coverage that can detect the exploitation of these vulnerabilities, download the latest rule sets from Snort.org, and our latest Vulnerability Advisories are always posted on Talos Intelligence’s website.  

**Denial-of-service, remote code execution vulnerabilities in TP-Link AC1350 router **

Talos researchers recently discovered four vulnerabilities in the TP-Link AC1350 wireless router. The AC1350 is one of many routers TP-Link produces and is designed to be used on home networks. 

TALOS-2023-1861 (CVE-2023-49074) is a denial-of-service vulnerability in the TP-Link Device Debug Protocol (TDDP). An attacker could exploit this vulnerability by sending a series of unauthenticated packets to the router, potentially causing a denial of service and forcing the device to reset to its factory settings.  

However, the TDDP protocol is only denial of serviceavailable for roughly 15 minutes after a device reboot.  

The TDDP protocol is also vulnerable to TALOS-2023-1862 (CVE-2023-49134 and CVE-2023-49133), a command execution vulnerability that could allow an attacker to execute arbitrary code on the targeted device. 

There is another remote code execution vulnerability, TALOS-2023-1888 (CVE-2023-49912, CVE-2023-49909, CVE-2023-49907, CVE-2023-49908, CVE-2023-49910, CVE-2023-49906, CVE-2023-49913, CVE-2023-49911) that is triggered if an attacker sends an authenticated HTTP request to the targeted device. This exploit includes multiple CVEs because an attacker could overflow multiple buffers to cause this condition. 

TALOS-2023-1864 (CVE-2023-48724) also exists in the device’s web interface functionality. An adversary could exploit this vulnerability by sending an unauthenticated HTTP request to the targeted device, thus causing a denial of service. 

**Multiple vulnerabilities in OAS Platform **

_Discovered by Jared Rittle._ 

Open Automation Software’s OAS Platform is an IoT gateway and protocol bus. It allows administrators to connect PLCs, devices, databases and custom apps. 

There are two vulnerabilities — TALOS-2024-1950 (CVE-2024-21870) and TALOS-2024-1951 (CVE-2024-22178) — that exist in the platform that can lead to arbitrary file creation or overwrite. An attacker can send a sequence of requests to trigger these vulnerabilities.  

An adversary could also send a series of requests to exploit TALOS-2024-1948 (CVE-2024-24976), but in this case, the vulnerability leads to a denial of service. 

An improper input validation vulnerability (TALOS-2024-1949/CVE-2024-27201) also exists in the OAS Engine User Configuration functionality that could lead to unexpected data in the configuration, including possible decoy usernames that contain characters not usually allowed by the software’s configuration. 

**Arbitrary write vulnerabilities in AMD graphics driver **

_Discovered by Piotr Bania._ 

There are two out-of-bounds write vulnerabilities in the AMD Radeon user mode driver for DirectX 11. TALOS-2023-1847 and TALOS-2023-1848 could allow an attacker with access to a malformed shader to potentially achieve arbitrary code execution after causing an out-of-bounds write. 

AMD graphics drivers are software that allows graphics processing units (GPUs) to communicate with the operating system.  

These vulnerabilities could be triggered from guest machines running virtualization environments to perform guest-to-host escape. Theoretically, an adversary could also exploit these issues from a web browser. Talos has demonstrated with past, similar, vulnerabilities that they could be triggered from HYPER-V guest using the RemoteFX feature, leading to executing the vulnerable code on the HYPER-V host.

Vulnerability in some TP-Link routers could lead to factory reset

Global organizations and geopolitical entities must adopt new strategies to combat the growing sophistication in attacks that parallel the complexities of our new geopolitical reality.

Source: Dragon Claws via Alamy Stock Photo

COMMENTARY

Today, it's rare for a month to pass without reports of new distributed denial-of-service (DDoS) attacks driven by geopolitical instability. Now, a single attack can include numerous countries and networks. While the war between Russia and Ukraine and elections in NATO countries like Poland drive geopolitically focused DDoS attacks, it's important to understand that these threats were present long before these conflicts, and will be around long after. But why are these attacks against governmental institutions so prominent, and why do they impact all of us in the first place? 

That is partially because changes in political leadership often correlate directly with a spike in DDoS attacks. For example, in Poland, DDoS attack volume increased fourfold within days of its new government being sworn into office. These spikes often result from hacktivist groups (e.g., KillNet, Noname057, Anonymous Sudan) opposing the viewpoints of newly elected officials. They impact the rest of us because DDoS attacks must cross multiple Internet service providers (ISPs) to reach the intended victim. 

Unfortunately, even an attack that is effectively mitigated will use up valuable resources on any ISP network it reaches. We refer to this as the "DDoS tax," which increases the cost of network operation, making it more expensive for everyone. In fact, global ISPs reported a 500% global increase in HTTP/S application layer attacks since 2019, and 17% growth in DNS reflection/amplification volumes in the first half of 2023. Furthermore, it is alarming that carpet-bombing attacks, a technique that simultaneously targets entire IP address ranges, increased by 110% from the first to the second half of 2022, with most attacks occurring against ISP networks.

The bottom line is that there is no escape from DDoS attacks on governmental institutions, and threat intelligence needs to be taken more seriously because of how universal the threat can be when it comes to compromising global ISP networks and additional IT infrastructure. Therefore, it is key for governmental entities to suppress DDoS attacks before they can begin.

**Comprehensive and Adaptive Defenses for Evolving DDoS Attacks**

With nation-states, bad actors often directly target Internet infrastructure to take out critical communications, ecommerce, and other vital infrastructure dependent on Internet connectivity. That means attackers target ISP networks to purposefully degrade Internet connectivity. Further, these bad actors typically have vastly more resources at their disposal than other attackers. They constantly innovate and explore new and more powerful DDoS attack vectors, evidenced by the creation of new ones every year, such as DNS Water Torture and Carpet Bombing. All the while, as DDoS defenses become more effective, bad actors continue to sidestep these defenses with new DDoS attack vectors and methodologies. These advanced techniques invariably find their way into the hands of criminal gangs and even individual hackers, who turn them against any entity from whom they can profit.

With the increased persistence of cybercriminals and the growth in complexity of DDoS attacks, the foundation for a comprehensive DDoS protection solution should identify and stop all types of DDoS attacks before they impact the availability of business-critical services. With today's growing frequency and complexity of DDoS attacks, a multilayer defense strategy is now no longer nice to have, but a requirement. New techniques such as adaptive DDoS strategies, which change vectors based on the defense that is presented, reinforce the need for better agility and efficiency when it comes to managing these increasingly complex attacks.

In the end, threat actors will continue using DDoS attacks as a way to further disrupt and inflame sociopolitical tensions around the world. Security professionals need to consider both local and international conflicts when assessing the DDoS risk factors associated with nation-state attacks. The unfortunate reality is that bad actors continue to find new ways to orchestrate attacks through evolving methodologies. As such, global organizations and geopolitical entities must adopt new strategies right now, such as advanced DDoS defense and suppression, to combat this growing sophistication in attacks that parallel the complexities of our new geopolitical reality.

**About the Author(s)**

Director, Security Solutions, Netscout

Gary is an industry veteran, bringing over 20 years of broad technology experience including routing and switching, wireless, mobility, collaboration, and cloud but always with a focus on security. His previous roles include solutions architect, security SME, sales engineering, consultancy, product management, IT and customer support. Prior to joining Netscout in 2012, he spent 12 years at Cisco Systems and held positions with Avaya and Cable & Wireless.

How Nation-State DDoS Attacks Impact Us All

By cybernewswire
Tel Aviv, Israel, April 10th, 2024, CyberNewsWire Cyber GRC software company Cypago has announced a new automation solution…
This is a post from HackRead.com Read the original post: Cypago Announces New Automation Support for AI Security and Governance

**Tel Aviv, Israel, April 10th, 2024, CyberNewsWire**

Cyber GRC software company Cypago has announced a new automation solution for artificial intelligence (AI) governance, risk management and compliance. This includes implementation of NIST AI RMF and ISO/IEC 42001, the newest AI security and governance frameworks. With more and more companies integrating new AI tools into their business processes, daily operations, and customer-facing products and services, safe and compliant use of AI has become a pivotal challenge.

Enterprises are adopting AI-powered tools and solutions at a staggering pace, thanks to the increasing power and democratization of AI platforms, and the extensive benefits that artificial intelligence can bring to business operations. However, AI also brings many dangers, including the potential for leaking private data, lack of transparency, and increasingly complex cyber threats. Organizations also need to prepare themselves for more legislation regulating the use and application of AI in business settings. 

The best way to protect against these dangers and stay ahead of AI regulations is through adherence to best practices for cyber GRC governance, which are in a constant state of flux. By providing comprehensive risk management, automated 24/7 monitoring, and cybersecurity governance tailored to AI solutions, Cypago enables companies to seamlessly secure their AI initiatives. 

“The world of AI is changing quickly, with new threats arising all the time and new regulations arriving frequently. We view it as our responsibility to help organizations maximize the benefits of AI while effectively mitigating the risks and ensuring compliance with best practices and good governance,” said Arik Solomon, CEO of Cypago. “These latest features ensure that Cypago supports the newest AI and cyber governance frameworks, enabling GRC and cybersecurity teams to automate GRC with the most up-to-date requirements.” 

Cypago delivers ongoing visibility into all the organization’s tools, applications, and models, while automating many processes needed for effective risk assessment and threat detection. Its heightened security measures for AI-based systems help keep data and software secure, reducing the risk of cyber threats, data breaches and regulatory violations. 

What’s more, Cypago is no stranger to the implementation of safe AI, having adopted natural language models and generative AI command prompts as part of its product in 2023.

The platform streamlines security, risk management and compliance gap detection and remediation processes, enabling teams to respond more effectively to emerging threats and vulnerabilities, while ensuring compliance with international, national, and industry-specific regulations. This empowers companies to navigate the complex compliance landscape, which now encompasses the use of AI, with confidence. 

**About Cypago**

Cypago’s revolutionary SaaS-based Cyber GRC Automation (CGA) platform redefines the three lines model by eliminating friction and bridging the gap between management, security, and operations. It transforms GRC initiatives into automated processes, enabling in-depth visibility, streamlining enforcement, and significantly reducing overall costs. The platform leverages innovative technologies, including advanced analysis and correlation engines, GenAI, and NLP models, designed to support any security framework in any IT environment, both in the cloud and on-premises. Cypago was founded in 2020 by tech leaders and cybersecurity veterans with decades of combined experience in the development, operations, and commercialization of cybersecurity solutions. For more information, visit https://cypago.com/. 

**Contact**

**PR**  
**Dan Edelstein**  
**InboundJunction**  
**\[email protected\]**

Cypago Announces New Automation Support for AI Security and Governance

Some companies let you opt out of allowing your content to be used for generative AI. Here’s how to take back (at least a little) control from ChatGPT, Google’s Gemini, and more.

If you’ve ever posted something to the internet—a pithy tweet, a 2009 blog post, a scornful review, or a selfie on Instagram—it has most likely been slurped up and used to help train the current wave of generative AI. Large language models, like ChatGPT, and image creators are powered by vast reams of our data. And even if it’s not powering a chatbot, the data can be used for other machine-learning features.

Tech companies have scraped vast swathes of the web to gather the data they claim is needed to create generative AI—with little regard for content creators, copyright laws, or privacy. On top of this, increasingly, firms with reams of people’s posts are looking to get in on the AI gold rush by selling or licensing that information. Looking at you, Reddit.

However, as the lawsuits and investigations around generative AI and its opaque data practices pile up, there have been small moves to give people more control over what happens to what they post online. Some companies now let individuals and business customers opt out of having their content used in AI training or being sold for training purposes. Here’s what you can—and can’t—do.

**There’s a Limit**

Before we get to how you can opt out, it’s worth setting some expectations. Many companies building AI have already scraped the web, so anything you’ve posted is probably already in their systems. Companies are also secretive about what they have actually scraped, purchased, or used to train their systems. “We honestly don't know that much,” says Niloofar Mireshghallah, a researcher who focuses on AI privacy at the University of Washington. “In general, everything is very black-box.”

Mireshghallah explains that companies can make it complicated to opt out of having data used for AI training, and even where it is possible, many people don’t have a “clear idea” about the permissions they’ve agreed to or how data is being used. That’s before various laws, such as copyright protections and Europe’s strong privacy laws, are taken into consideration. Facebook, Google, X, and other companies have written into their privacy policies that they may use your data to train AI.

While there are various technical ways AI systems could have data removed from them or “unlearn,” Mireshghallah says, there’s very little that’s known about the processes that are in place. The options can be buried or labor-intensive. Getting posts removed from AI training data is likely to be an uphill battle. Where companies are starting to allow opt-outs for future scraping or data sharing, they are almost always making users opt-in by default.

“Most companies add the friction because they know that people aren’t going to go looking for it,” says Thorin Klosowski, a security and privacy activist at the Electronic Frontier Foundation. “Opt-in would be a purposeful action, as opposed to opting out, where you have to know it’s there.”

While less common, some companies building AI tools and machine learning models don't automatically opt-in customers. “We do not train our models on user-submitted data by default. We may use user prompts and outputs to train Claude where the user gives us express permission to do so, such as clicking a thumbs up or down signal on a specific Claude output to provide us feedback,” says Jennifer Martinez, a spokesperson for Anthropic. In this situation, the most recent iteration of the company’s Claude chatbot is built on public information online and third-party data—content people posted elsewhere online—but not user information.

The majority of this guide deals with opt-outs for text, but artists have also been using “Have I Been Trained?” to signal their images shouldn't be used for training. Run by startup Spawning, the service allows people to see if their creations have been scraped and then opt out of any future training. “Anything with a URL can be opted out. Our search engine only searches images, but our browser extension lets you opt out any media type,” says Jordan Meyer, cofounder and CEO of Spawning. Stability AI, the startup behind a text-to-image tool called Stable Diffusion, is among companies that say they are honoring the system.

The list below only includes companies currently with an opt-out process. For example, Microsoft’s Copilot does not offer users with personal accounts the option to have their prompts not used to improve the software. “A portion of the total number of user prompts in Copilot and Copilot Pro responses are used to fine-tune the experience,” says Donny Turnbaugh, a spokesperson for Copilot. “Microsoft takes steps to deidentify data before it is used, helping to protect consumer identity.” Even if the data is deidentified, privacy-minded users may want more potential control over their information.

**How to Opt Out of AI Training**

Adobe

Adobe via Matt Burgess

If you store your files in Adobe’s Creative Cloud, the company may use them to train its machine-learning algorithm. “When we analyze your content for product improvement and development purposes, we first aggregate your content with other content and then use the aggregated content to train our algorithms and thus improve our products and services,” reads the company’s FAQ. This doesn’t apply to any files stored only on your device.

If you’re using a personal Adobe account, it’s easy to opt out. Open up Adobe’s **privacy page**, scroll down to the **Content analysis** section, and click the toggle to turn it off. For business or school accounts, the opt-out process is not available on the individual level, and you’ll have to reach out to your administrator.

Amazon: AWS

AI services from Amazon Web Services, like Amazon Rekognition or Amazon CodeWhisperer, may save customer data to improve the company’s tools. Head’s up, this is the most complicated opt-out process included in the roundup, so you likely need help from an IT professional at your company or an AWS representative to perform it successfully. Outlined on this support page from Amazon, the process includes enabling the option for your organization, creating a policy, and attaching that policy where necessary.

Google: Gemini

For users of Google’s chatbot, Gemini, conversations may sometimes be selected for human review to improve the AI model. Opting out is simple, though. Open up Gemini in your browser, click on **Activity**, and select the **Turn Off** drop-down menu. Here you can just turn off the Gemini Apps Activity, or you can opt out as well as delete your conversation data. While this does mean in most cases that future chats won’t be seen for human review, already selected data is not erased through this process. According to Google’s privacy hub for Gemini, these chats may stick around for three years.

Grammarly

Grammarly does not currently offer an opt-out process for personal accounts, but self-serve business accounts can choose to opt out from having their data used to train Grammarly’s machine-learning model. Turn it off by opening up your **Account Settings**, clicking on the **Data Settings** tab, and toggling off **Product Improvement & Training**. If you have a managed business account, which includes accounts for classroom education and accounts bought through a Grammarly sales representative, you are automatically opted out from AI model training.

HubSpot

HubSpot, a popular marketing software, automatically uses data from customers to improve its machine-learning model. Unfortunately, there’s not a button to press to turn off the use of data for AI training. You have to **send an email** to privacy@hubspot.com with a message requesting that the data associated with your account be opted out.

OpenAI: ChatGPT and Dall-E

OpenAI via Matt Burgess

People reveal all sorts of personal information while using a chatbot. OpenAI provides some options for what happens to what you say to ChatGPT—including allowing its future AI models not to be trained on the content. “We give users a number of easily accessible ways to control their data, including self-service tools to access, export, and delete personal information through ChatGPT. That includes easily accessible options to opt out from the use of their content to train models,” says Taya Christianson, an OpenAI spokesperson. (The options vary slightly depending on your account type, and data from enterprise customers is not used to train models).

On its help pages, OpenAI says ChatGPT web users without accounts should navigate to **Settings** and then uncheck **Improve the model for everyone**. If you have an account and are logged in through a web browser, select **ChatGPT, Settings, Data Controls,** and then turn off **Chat History & Training**. If you’re using ChatGPT’s mobile apps, go to **Settings**, pick **Data Controls,** and turn off **Chat History & Training**. Changing these settings, OpenAI’s support pages say, won’t sync across different browsers or devices, so you need to make the change everywhere you use ChatGPT.

OpenAI is about a lot more than ChatGPT. For its Dall-E 3 image generator, the startup has a **form that allows you to send images** to be removed from “future training datasets.” It asks for your name, email, whether you own the image rights or are getting in touch on behalf of a company, details of the image, and any uploads of the image(s). OpenAI also says if you have a “high volume” of images hosted online that you want removed from training data, then it may be “more efficient” to add GPTBot to the robots.txt file of the website where the images are hosted.

Traditionally a website’s robots.txt file—a simple text file that usually sits at websitename.com/robots.txt—has been used to tell search engines, and others, whether they can include your pages in their results. It can now also be used to tell AI crawlers not to scrape what you have published—and AI companies have said they’ll honor this arrangement.

Perplexity

Perplexity is a startup that uses AI to help you search the web and find answers to questions. Like all of the other software on this list, you are automatically opted in to having your interactions and data used to train Perplexity’s AI further. Turn this off by clicking on your **account name**, scrolling down to the **Account** section, and turning off the **AI Data Retention** toggle.

Quora

Quora via Matt Burgess

Quora says it “currently” doesn’t use answers to people’s questions, posts, or comments for training AI. It also hasn’t sold any user data for AI training, a spokesperson says. However, it does offer opt-outs in case this changes in the future. To do this, visit its **Settings** page, click to **Privacy,** and turn off the “**Allow large language models to be trained on your content**” option. Despite this choice, there are some Quora posts that may be used for training LLMs. If you reply to a machine-generated answer, the company’s help pages say, then those answers may be used for AI training. It points out that third parties may just scrape its content anyway.

Rev

Rev, a voice transcription service that uses both human freelancers and AI to transcribe audio, says it uses data “perpetually” and “anonymously” to train its AI systems. Even if you delete your account, it will still train its AI on that information.

Kendell Kelton, head of brand and corporate communications at Rev, says it has the “largest and most diverse data set of voices,” made up of more than 6.5 million hours of voice recording. Kelton says Rev does not sell user data to any third parties. The firm’s terms of service say data will be used for training, and that customers are able to opt out. People can opt out of their data being used by **sending an email** to support@rev.com, its help pages say.

Slack

All of those random Slack messages at work might be used by the company to train its models as well. “Slack has used machine learning in its product for many years. This includes platform-level machine-learning models for things like channel and emoji recommendations,” says Jackie Rocca, a vice president of product at Slack who’s focused on AI.

Even though the company does not use customer data to train a large language model for its Slack AI product, Slack may use your interactions to improve the software’s machine-learning capabilities. “To develop AI/ML models, our systems analyze Customer Data (e.g. messages, content, and files) submitted to Slack,” says Slack’s privacy page. Similar to Adobe, there’s not much you can do on an individual level to opt out if you’re using an enterprise account.

The only real way to opt out is to have your administrator email Slack at feedback@slack.com. The message must have the subject line “Slack Global model opt-out request” and include your organization's URL. Slack doesn’t provide a timeline for how long the opt-out process takes, but it should send you a confirmation email after it’s complete.

Squarespace

Website-building tool Squarespace has built in a toggle to stop AI crawlers from scraping websites it hosts. This works by updating your website’s robots.txt file to tell AI companies the content is off limits. To block the AI bots, open **Settings** within your account, find **Crawlers**, and turn off **Artificial Intelligence Crawlers**. It points out this should work for the following crawlers: Anthropic, OpenAI’s GPTBot and ChatGPT-User, Google Extended, and CCBot.

Substack

If you use Substack for blog posts, newsletters, or more, the company also has an easy option to apply the robots.txt opt-out. Within your **Settings** page, scroll to the **Publication** section and turn on the toggle to **Block AI training**. Its help page points out: “This will only apply to AI tools that respect this setting.”

Tumblr

Blogging and publishing platform Tumblr—owned by Automattic, which also owns WordPress—says it is “working with” AI companies that are “interested in the very large and unique set of publicly published content” on the wider company’s platforms. This doesn’t include user emails or private content, an Automattic spokesperson says.

Tumblr has a “prevent third-party sharing” option to stop what you publish being used for AI training, as well as being shared with other third parties such as researchers. If you’re using the Tumblr app, go to account S**ettings**, select your blog, click on the **gear icon**, select **Visibility**, and toggle the “**Prevent third-party sharing**” option. Explicit posts, deleted blogs, and those that are password-protected or private, are not shared with third-party companies in any case, Tumblr’s support pages say.

WordPress

Wordpress via Matt Burgess

Like Tumblr, WordPress has a “prevent third-party sharing” option. To turn this on, visit your website’s dashboard, click on **Settings**, **General**, and then through to **Privacy**, select the **Prevent third-party sharing** box. “We are also trying to work with crawlers (like https://commoncrawl.org/) to prevent content from being scraped and sold without giving our users choice or control over how their content is used,” an Automattic spokesperson says.

Your Website

If you are hosting your own website, you can update your robots.txt file to tell AI bots not to scrape the pages. Most news websites don’t allow their articles to be crawled by AI bots. WIRED’s robots.txt file, for example, doesn’t allow bots from OpenAI, Google, Amazon, Facebook, Anthropic, or Perplexity, among others. This opt-out isn’t just for publishers though: Any website, big or small, can alter its robots file to exclude AI crawlers. All you need to do is add a disallow command; working examples can be found here.

How to Stop Your Data From Being Used to Train AI

Cybersecurity researchers have disclosed what they say is the "first native Spectre v2 exploit" against the Linux kernel on Intel systems that could be exploited to read sensitive data from the memory.
The exploit, called Native Branch History Injection (BHI), can be used to leak arbitrary kernel memory at 3.5 kB/sec by bypassing existing Spectre v2/BHI mitigations, researchers from Systems and

Researchers Uncover First Native Spectre v2 Exploit Against Linux Kernel

Google has integrated Mandiant's security offerings into its AI platform to detect, stop, and remediate cybersecurity attacks as quickly as possible.

Source: Klaus Ohlenschlaeger via Alamy Stock Photo

Gemini now has security capabilities: Google has integrated Mandiant's security offerings into its artificial intelligence (AI) platform, the company announced during its Google Cloud Next conference in Las Vegas.

The new capabilities are automated security agents that use generative AI to detect, stop, and remediate cybersecurity attacks as quickly as possible and to to analyze code to find security problems. The agents will assist security operations teams by increasing the speed of investigations.

"Security agents will help you in every stage of the security life cycle, from prevention, detection and response," said Google Cloud CEO Thomas Kurian during a keynote speech at the event.

The security offerings are built around Google's Gemini large language model, which is also being used to summarize documents and generate images. The new security features use generative AI to analyze code to find security problems. Security analysts can run deep queries to get to the root cause of problems and seek answers. The technology will also better explain findings so companies can take action to neutralize threats.

Google is betting its future on AI products such as Gemini, which it introduced earlier this year. Gemini is already being integrated into search and productivity applications. Mandiant's offerings include threat detection and security operations, and Gemini adds AI capabilities to cover detection, analysis, and prevention.

**Using AI to Understand Attacks**

One of the new products, Gemini in Threat Intelligence, uses natural language prompts to get deep insight about malicious behavior.

"We're able to take the experience and intelligence we gather from protecting Google's own services, combine it with Mandiant's leading, frontline insight from their work in incident response to show emerging threats, the severity, and risk factors," Kurian said.

The intelligent threat offers ties into the next product, Gemini in Security Operations, which relies on AI to improve the speed to address threats from malicious behavior. It can also be used to summarize and explain findings, recommend next steps, and put remediation playbooks in action.

Gemini can recommend actions based on factors such as detection rules, and provide recommendations for faster response times.

"Analysts can now ask Gemini for the latest threat intelligence from Mandiant directly in-line — including any indicators of compromise found in their environment — and Gemini will navigate users to the most relevant pages in the integrated platform for deeper investigation," Google wrote in a blog post.

The offering uses Gemini 1.5 Pro's AI capabilities to analyze large samples of potential malicious code or malware, the interaction between modules, and to reveal their true malicious intent, Kurian said.

The final product announced was Gemini in Security Command Center, which evaluates a security posture, provides findings, and summarizes attack paths, which helps companies remediate cloud risks.

"We've leveraged tremendous amounts of machine learning, precision AI, generative AI, to make sure our customers are able to detect, stop, and remediate all cybersecurity attacks as quickly as possible," Kurian said.

AI tools such as Gemini and OpenAI's ChatGPT are improving productivity, but hallucinations and other shortcomings are raising concerns about the technology being used irresponsibly. The White House late last month issued an executive order on responsible use and development of AI development, which also recommends monitoring AI systems to prevent misuse.

**About the Author(s)**

Agam Shah has covered enterprise IT for more than a decade. Outside of machine learning, hardware, and chips, he's also interested in martial arts and Russia.

Google Gives Gemini a Security Boost

PRESS RELEASE

WEST PALM BEACH, Fla., April 08, 2024 (GLOBE NEWSWIRE) -- Veriato released their next generation Insider Risk Management (IRM) solution, re-imagined from the ground up to deliver superior insider risk management and threat detection. With organizations of all sizes facing a more complex cybersecurity environment, Veriato IRM delivers unprecedented flexibility and scalability using the power of GenAI.

Veriato's IRM solution offers best-in-class technology for companies looking to improve their threat mitigation with AI enabled predictive analytics delivering better detection and predictability. According to IBM, the average cost of a single data breach has reached $4.45 million, a record high. Organizations need a holistic view of employee behavior to get ahead of threats and to reduce their exposure and costs. The new platform provides the agility and usability that make it a must-have in the fight against increasing insider threats. Features such as predictive risk intelligence, risk scoring, baselining and PII identification offer market-leading detection capabilities that are easy to deploy and manage.

Veriato IRM delivers:

*   Unparalleled risk detection capabilities with Generative AI.
    
*   Advanced behavior pattern identification across logon events, document activity, email activity, etc.
    
*   Complex dimensional analysis, including Risk Scoring.
    
*   Comprehensive language and sentiment analysis.
    
*   Automatic PII/PHI identification with RAG pre-trained models.
    
*   Flexible and scalable microservices architecture.
    
*   Multiple deployment options – cloud, on-premise, hybrid.
    

Veriato IRM is the product of close collaboration with Veriato’s global base of customers across a number of industries including financial services, healthcare and government. Designed with a deep understanding of today’s changing workforce and the technology and data needs of today’s modern organization, Veriato IRM is the most advanced and usable solution on the market today.

“Insider Risk Management has become essential for organizations committed to protecting customer data, IP and their employee bases now that the future of work is here. From board rooms to SMBs, IRM is at the top of security agendas. Applying behavioral analytics to cyber allows leaders to get ahead of breaches instead of just reacting to them when they occur,” said Elizabeth Harz, CEO of Veriato. “We are extremely excited to build upon our history as the category creator, and provide a new layer of control, transparency, confidence and perhaps most importantly, proactivity.”

About Veriato

Veriato is reinventing the category it created, using AI-based user behavior analytics to help companies prevent risks and increase productivity in their remote, hybrid and in-office environments. Veriato’s platform offers solutions for Insider Risk Management (IRM), behavioral analytics, user activity monitoring (UAM) and data loss prevention (DLP) in a single powerful platform. Veriato delivers monitoring, salient alerts, reporting and screenshots, allowing customers to be predictive and proactive rather than reactive, critical in cybersecurity. The platform helps global Enterprises, SMBs and Government entities become more engaged, productive and secure.

You May Also Like

Veriato Launches Next Generation Insider Risk Management Solution

The US Congress will this week decide the fate of Section 702, a major surveillance program that will soon expire if lawmakers do not act. WIRED is tracking the major developments as they unfold.

The United States government, like its rivals in Moscow and Beijing, has poured untold millions of dollars into quietly turning the phones and internet browsers of its own citizens into a powerful intelligence-gathering tool. Shadowy deals between federal agencies and commercial data brokers have helped the US intelligence system to amass a “large amount” of what its own experts term “intimate information” on Americans.

Most US citizens are in the dark as to the true scope and scale of the surveillance they’re under.

House speaker Mike Johnson—whose brief tenure as speaker has been roiled by an ongoing debate over domestic intelligence abuses—previously supported several privacy measures that, now in power, he is working to defeat, including strong new limits on the government’s access to private data.

This week, Johnson is working to resolve the lingering issue of reauthorizing Section 702, a key foreign surveillance program authorized by Congress to target terrorists, cybercriminals, and narcotics traffickers overseas. The program is set to officially expire on April 19.

Congressional sources tell WIRED that a vote to salvage the program could come as early as Thursday, following a series of scheduled briefings on Tuesday and Wednesday between lawmakers and intelligence officials, as well as a number of smaller votes that may significantly modify the terms of the program for years to come.

The focus of privacy advocates has turned almost entirely to an amendment that aims to force the FBI and other agencies to apply for a warrant before accessing the communications of Americans _incidentally_ captured by the US under the 702 program.

Thursday’s vote over the 702 program is at least the third scheduled by the speaker since December. As Johnson has in each case waited until the final moment to delay the vote, a fog of uncertainty surrounds the whole of the process. Privately, lawmakers are discussing next steps should the speaker decide to simply let the program expire, avoiding new statutory limits on the government's most-prized surveillance weapon.

To keep pace with a situation that is sure to evolve rapidly over the next 48 hours, WIRED will update this article with the latest details as they become available. See below for the latest developments.

**House Intel Chair Fires Back at Warrant Seekers**

“We are here,” representative Mike Turner told the House Rules Committee, “because the intelligence community failed us. The FBI failed us. The intelligence community failed to properly police themselves, and they abused the tools that we gave them under FISA.”

Turner, the GOP chairman of the House Intelligence Committee, testified that, in his opinion, the underlying text of the bill was adequate to curb future surveillance abuses by the FBI; that the abuses of the past had been the result of mistakes made by “rank-and-file” employees; and that the same mistakes could be reasonably avoided today by merely asking higher ranking officers to sign off before granting access to wiretaps of Americans’ calls and texts.

Throughout his testimony, Turner discussed the 702 program in terms that focused strictly on its use against terrorist organizations, at times implying that only the communications of Americans in direct contact with foreigners who “represent a national security threat to the United States” are swept up by the program.

In fact, no foreigner is required to have ties to terrorism or be suspected of a crime against the US to be considered a legitimate 702 target. Far less discerning, one of the stated purposes of the program is to gather information relevant to the “conduct of foreign affairs”—a phrase that legal experts contend is so vague as to apply to “almost any activity.”

**What Is Section 702?**

In the wake of 9/11, US president George W. Bush authorized the National Security Agency (NSA) to eavesdrop on Americans without court-approved warrants as part of the hunt for evidence of terrorist activity. A federal judge ruled the collection unconstitutional in 2006, as part of a lawsuit brought by the American Civil Liberties Union. (An appeals court later overturned the ruling without challenging the case's merits.)

Rather than end the surveillance, Congress codified the program as Section 702 of the Foreign Intelligence Surveillance Act (FISA), granting itself some authority to enforce procedures ostensibly designed to limit the program's impact on Americans’ civil liberties.

Section 702 explicitly prohibits the government from targeting Americans. The surveillance must instead focus on foreigners who are physically located overseas. Nevertheless, Americans’ communications are routinely swept up by the program.

While denying that it intentionally sets out to eavesdrop on its own citizens, once it has already done so, the US government’s position is that it now has a right to access these “legally collected” communications without a judge’s approval. In 2021 alone, the FBI conducted searches of communications intercepted under 702 more than 3.4 million times.

Last year, after acknowledging that hundreds of thousands of these searches were unlawful, the FBI said it had taken steps to curtail the number of queries carried out by its employees, reporting in 2022 as few as 204,000 searches.

It is impossible to count the number of Americans whose calls, emails, and texts are subject to surveillance under 702, the government claims, arguing that any attempt to reach an accurate figure would only further imperil the privacy of the Americans it surveils.

Congress is currently divided into two factions: Those that believe the FBI should be required to get a warrant before reading or listening to the communications of Americans collected under 702. And those who say warrants are too burdensome a requirement to impose on investigations of national security threats.

**Judiciary Leaders Call Out FBI’s History of Abuse**

Jim Jordan, the Republican chairman of the House Judiciary Committee, and a leading advocate for stronger privacy under 702 among conservatives, drew attention in his testimony Tuesday to the litany of surveillance abuses first reported by _The Washington Post_ one year ago.

Court documents declassified and released by the FISA court last year found that the FBI had misused the 702 program more than 278,000 times, including, as _The Post_ reported, against “crime victims, Jan. 6 riot suspects, people arrested at protests after the policing killing of George Floyd in 2020 and—in one case—19,000 donors to a congressional candidate.”

“The fundamental question, still, is if the FBI wouldn't follow the procedures in place 278,000 times ... are new requirements going to be enough to safeguard American’ liberties?” Jordan says. "When you have the history we have with this organization, relative to not following the rules, we think you need a separate but equal branch of the government to approve a warrant before you can query American citizens' information.”

Jarrold Nadler—Jordan's Democratic counterpart on the Judiciary—concurred, calling even more emphatically on Congress to “rein in abuse of FISA Section 702 authorities by the intelligence community.”

“Yes, there are already laws on the books to protect Americans from unauthorized government surveillance,” says Nadler. “But we know from our oversight efforts and the intelligence community's own reporting that these laws are often disregarded and are at the very least inadequate to keep this powerful surveillance tool in check.”

The House Rules Committee is currently holding a hearing that will determine what the final text of the 702-reauthorization bill will look like and which amendments will be offered up for a vote on the House floor.

Both the House Judiciary and Intelligence committees are expected to offer up several amendments, with the latter aiming to impose new warrant requirements on the FBI prior to querying the 702 database for information on Americans. Watch the hearing below:

Tag

#intel