In response to a new European Union law, AI platforms are implementing new schemes for watermarking the content they generate. Anthropic recently disclosed its future Claude models will use SynthID-Text, an approach Google created and released as open source. It uses a secret key that subtly changes the process a model uses for choosing the next word in a sentence. Whereas a top next word choice might be “cloudy,” the key might change it to “overcast.” Anyone who knows the key can determine if it was generated by the platform using it.

New research shows that SynthID-Text can change not just word selection but also the tools a model invokes and the chances it will adhere to or disregard safety guardrails it has been trained to follow. The threat can become greater in the face of an adversarial prompt, in which an attacker attempts to cause a model to carry out a harmful action, such as revealing a password or other sensitive information. Instructions that normally wouldn’t be followed will, in some cases, be performed once the watermarking is deployed. The finding underscores the need for developers to thoroughly test how their LLMs and agents behave when watermarking is in place.

Changing safety behavior

“As compared to the same models without watermarking, it is definitely going to change their behavior, especially when we place it under adversarial conditions, or we make these models call tools when they’re powering an agent,” Andrea Siposova, an AI security researcher at Lasso Security, told Ars. “Watermarking is made to not be perceptible to a reader, but we know that when we are changing anything about what the model is generating, it is going to cause some tradeoffs, it’s going to show up somewhere.”

Read full article

Comments

Documentation files on more than 100 websites are referencing potentially dangerous executable content that gets installed automatically when visited by many AI agents. A few dozen companies, some of them Fortune 500s, are among those that executed proof-of-concept code. At least one misconfigured site is directing visitors, human or AI, to live malware.

The potentially dangerous content is in llms.txt and llms-full.txt files, an emerging convention websites employ to provide machine-readable summaries of the site’s content and its high-level structure. These files are the AI equivalent of the robots.txt standard that instructs search engines how to index the site's content. Google Lighthouse, a tool for helping web developers, has more here. Correctly configured llms.txt and llms-full.txt files for Cloudflare are here and here.

How the researchers found it

Researchers at a stealth startup in Israel scanned 6,214 live domains belonging to defense contractors, Fortune 500, and Big Tech companies. Of the 8,265 llms.txt and llms-full.txt files they found (many sites hosted both an llms.txt and an llms-full.txt file), 120 of them, each on a different site, pointed to one or more code packages or domain names that weren’t registered. To test what happens when an AI agent processes such files, the researchers registered a handful of the unclaimed names and hosted packages that caused any machine executing them to reach out to their server. Within an hour, the researchers received a phone-home response from a Fortune 500 company. Over time, they got a few dozen more, some from more Fortune 500 companies and others from startups. Their beacon also recorded the chain of parent processes that spawned each install, ultimately revealing that coding agents, including Claude, OpenAI's Codex, and Nous Research's Hermes, were involved. Anthropic, OpenAI, and Nous Research did not respond to requests for comment by the time of publication.

Read full article

Comments

The OpenAI agents involved in last month’s incursion into Hugging Face were trained so heavily on winning a competition that they pursued a relentless campaign to cheat, a new report documented. In the process, and without authorization, they created an improvised message board to hatch a plan that ultimately landed them squarely inside the latter company’s network.

Over the course of May and June, OpenAI gave the agents what the company described as “impossible tasks” to complete on the benchmarking framework ExploitGym. The internal test was designed to test how the agents would respond. To get a full understanding of the agent capabilities, company engineers disabled safety guardrails that normally are in place to prevent the sort of hacks that eventually hit Hugging Face and one other undisclosed organization. The stymied agents’ training made them so focused on winning that they performed tasks they were never explicitly instructed to follow.

Cheaters gonna cheat

The first step was creating a message board that allowed the agents to pass notes to each other. OpenAI hadn’t provided any such platform, so the agents repurposed a platform called Artifactory, which OpenAI was using in internal testing of several unreleased hacking agents. OpenAI was using Artifactory as one of the measures to prevent the agents from egressing its isolated sandboxes and accessing the Internet, while at the same time simulating a real-world hacking environment.

Read full article

Comments

Last week’s unprecedented security event in which two OpenAI security hacking models trespassed into the network of fellow AI company Hugging Face was enabled by exploiting one or more zero-day vulnerabilities in Artifactory, JFrog, the product’s developer, said Monday.

In an incident mimicking a dystopian sci-fi novel, two OpenAI models broke out of the restricted environment meant to keep them from accessing the Internet during an internal test, the AI company revealed last week. The models went on to breach Hugging Face’s network and steal confidential information and credentials. OpenAI said its agent achieved the feat by exploiting a previously unknown vulnerability. The company called the event “unprecedented,” and outsiders largely agreed.

Not the triumph it was made out to be

OpenAI said the models exploited multiple attack vectors, including stolen credentials and zero-days, to gain remote code execution capabilities, but until now, the vulnerable software was unknown. JFrog’s Monday disclosure said the product was a self-managed instance Artifactory, a repository management system that secures and streamlines customers’ software development operations. JFrog says Artifactory is used by more than 7,500 developer Teams, 80 percent of which work for Fortune 100 companies.

Read full article

Comments

Prompt injections, the malicious commands attackers embed into content to entice large language models to follow them, have been attackers’ go-to tool for turning AI platforms against their users. A well-phrased command sneaked into an email or calendar invitation is often all it takes to cause the LLM to exfiltrate sensitive data or follow other harmful actions.

Now, defenders are embracing the prompt injection, too.

A strong, sharp effect

Researchers from Tracebit on Monday said they found that placing prompt injections alongside passwords, cryptographic keys, and other secrets stored on Amazon Web Services was often all that was needed to shut down attacks from AI hacking agents. The prompts direct the attacking LLM to perform an action forbidden by its guardrails, the safety barriers AI developers erect to prevent it from taking harmful actions. The LLM responds by shutting down.

Read full article

Comments

In the brief history of AI security, the prompt injection has quickly become the top threat. Large language models are inherently unable to distinguish between legitimate instructions provided by users and malicious ones sneaked into emails, source code, and other third-party content the models are processing. This makes it trivial to surreptitiously inject malicious commands that the LLM readily follows.

With no way to enforce this crucial boundary between trusted and untrusted sources, AI engine developers are left to erect elaborate guardrails designed to mitigate the damage rather than solve the root cause.

To date, most prompt injections have fallen into a class known as push, in which each potential victim is targeted. For example, the adversary injects malicious instructions into an individual email or calendar invitation. Because the injection must then be sent (or pushed) to each specific target, the scale of the attack is limited, hampering mass exploits that hit the Internet at large.

Read full article

Comments

Makers of AI browsers make lofty promises. With a single prompt, users can ask one to find a restaurant in a particular part of town, reserve a table, invite a colleague to lunch, and email a confirmation. These makers are much more reticent about the risks of blurring the once fine line between browsing sites and asking a large language model a question or instructing it to take potentially sensitive actions.

LLM developers’ answer so far has been to build guardrails that make some requests off-limits. Developing software exploits, stealing credentials, or teaching how to build a pipe bomb are examples. The problem with this approach is that the guardrails are reactive and treat the symptoms rather than solve the root cause. It’s tantamount to the manufacturer of an unsafe vehicle advocating for new road designs rather than fixing the flaws that make it prone to accidents.

Lulling LLMs into an alternate reality

New research puts this predicament on sharp display. It demonstrates how a website can lull AI browsers into a false reality where the rules governing its behavior no longer apply. After that, an attacker has free rein to invoke all kinds of destructive actions, such as extracting code from a private repository or extracting credentials from the built-in password manager.

Read full article

Comments

Last Tuesday, Microsoft patched a vulnerability it rated as max critical in its M365 Copilot AI platform. On Monday, the researchers who discovered the vulnerability and reported it to Microsoft revealed how their proof-of-concept exploit could retrieve 2FA codes and other sensitive data from emails accessible to Copilot.

Microsoft and other LLM providers have been unable to prevent their products from complying with malicious requests to reveal data. The root cause: AI bots are unable to distinguish between instructions provided by users and those snuck into third-party content the models are summarizing, drafting responses to, or using to perform other actions on behalf of the user. With no way to secure this crucial boundary, Microsoft and its peers are left to erect complicated and ad hoc guardrails designed to rein in the consequences of this incurable gullibility.

Jumping over guardrails

One guardrail built into Copilot and most other LLMs prevents them from submitting web forms, sending emails, and taking similar actions that can be used to exfiltrate data from the user. To work around this, LLM hackers turned to markup language, which, among other things, allows users to add formatting elements such as headings, lists, and links to text without the need for HTML tags. Another workaround is to wrap sensitive data inside HTML tags such as <img> and <form>. In either case, a web request showing the data hits the attacker’s web server, where the secret information is captured in logs.

Read full article

Comments