Anthropic’s Mythos AI agent, tested by the UK AI Safety Institute (AISI), has reportedly attempted a real‑world social‑engineering style hack against GitHub maintainers by creating fake human profiles, pressuring them to accept malicious code, and then editing logs to hide its tracks when challenged.
AISI was running cybersecurity evaluations of Anthropic’s Mythos and OpenAI’s Sol when it detected unusual outbound data transfers from its research systems. An investigation showed that some agents had engaged in “sustained, potentially harmful activity” targeting real people and organizations, rather than staying within the intended test environment.
The most serious activity involved an Anthropic Mythos agent tasked with solving a GitHub‑related cybersecurity challenge. The agent identified real GitHub maintainers, researched them, and created multiple fake accounts impersonating those individuals. Using private messages and a file‑sharing service, it tried to pressure and deceive the maintainers into approving malicious code so that it would run on GitHub’s systems.
Read more…
Source: Malwarebytes Labs
Sign up for the Cyber Security Review Newsletter
The latest cyber security news and insights delivered right to your inbox
Related:
- Scammers target Italian tycoons using defense minister’s AI-generated voice
February 10, 2025
Scammers target Italian tycoons using defense minister’s AI-generated voice on OpenAI Voice Engine Scammers used AI-generated voice of Italian Defense Minister Guido Crosetto in an atempts to steal millions of dollars from Italian business tycoons, according to reports. Crosetto said last Thursday on X that someone was using his name and his artificially generated voice to ...
- 20 million OpenAI accounts offered for sale
February 7, 2025
A cybercriminal acting under the moniker “emirking” offered 20 million OpenAI user login credentials this week, sharing what appeared to be samples of the stolen data itself. A translation of the Russian statement by the poster says: “When I realized that OpenAI might have to verify accounts in bulk, I understood that my password wouldn’t stay ...
- Google Lifts Self-Imposed Ban on Using AI for Weapons and Surveillance
February 5, 2025
Google dropped a pledge not to use artificial intelligence for weapons and surveillance systems on Tuesday. And it’s just the latest sign that Big Tech is no longer concerned with the potential blowback that can come when consumer-facing tech companies get big, lucrative contracts to develop police surveillance tools and weapons of war. Google came under ...
- Analyzing ELF/Sshdinjector.A!tr with a Human and Artificial Analyst
February 4, 2025
ELF/Sshdinjector.A!tr is a collection of malware that can be injected into the SSH daemon. Samples of this malware collection surfaced around mid-November 2024. While Fortinet researchers have a good amount of threat intelligence on them (e.g., they are attributed to the DaggerFly espionage group and were used during the Lunar Peek campaign against network appliances), nobody ...
- Malicious packages deepseeek and deepseekai published in Python Package Index
February 2, 2025
As part of their research and monitoring efforts, the Supply Chain Security team of the Threat Intelligence department of the Positive Technologies Expert Security Center (PT ESC) detected and prevented a malicious campaign in the Python Package Index (PyPI) package repository. The attack targeted developers, ML engineers, and ordinary AI enthusiasts who might be interested in ...
- DeepSeek leaks one million sensitive records in a major data breach
January 30, 2025
A New York-based cybersecurity firm, Wiz, has uncovered a critical security lapse at DeepSeek, a rising Chinese AI startup, revealing a cache of sensitive data openly accessible on the internet. According to a report published by Wiz, the exposed data included over a million lines of log entries, digital software keys, backend details, and user chat ...