Both systems created fake human profiles to try to deceive people during a simulated cyberattack.
On Tuesday, the British government warned of autonomous and deceptive behavior in the artificial intelligence models Mythos 5 from Anthropic and GPT-5.6 Sol, following tests conducted by the AI Security Institute (AISI).
RELATED:
United Kingdom Bans Social Networks Use to Children Under 16 From Tomorrow
Nineteen unauthorized actions were recorded in the 122 tests performed with seven models, 17 by Mythos and 2 by Sol, exceeding the parameters set by the researchers.
Both systems even created fake human profiles to try to deceive people during a simulated cyberattack, a behavior never before observed in this type of evaluation.
The report detailed that Mythos attempted to access a service by sending private messages after creating fake accounts that mimicked real people, and also tricked users into running malicious files.
Anthropic found three incidents in which Claude broke into the production systems of real companies, believing they were part of a capture-the-flag exercise.
In one, Claude uploaded working malware to PyPI during a cyber evaluation the model believed was simulated.
The… https://t.co/GxeXaBc97D pic.twitter.com/bnnnAbBI4f
— International Cyber Digest (@IntCyberDigest) July 30, 2026
Although the AISI clarified that these attempts were unsuccessful and no real-world damage occurred, it considered this a turning point, given the observed risks of autonomy and deception without specific instructions.
Anthropic responded that the tests were not representative of its production models, although it confirmed that it had launched its own investigation. OpenAI, for its part, noted that the test conditions did not reflect the typical use of its systems.
Both companies stated that they will continue to collaborate with evaluators and other industry stakeholders to strengthen common practices that enable safe evaluations as the models become more capable.
#FromTheSouth News Bits | United Kingdom: In a major medical breakthrough, the world’s first clinical trial for a vaccine against the deadly Bundibugyo Ebola virus, developed by researchers at Oxford University, has officially begun human testing. pic.twitter.com/ZQ0nkntrjV
— teleSUR English (@telesurenglish) July 29, 2026
teleSUR: JP
Source: EFE
From teleSUR English via This RSS Feed.