Routine cybersecurity testing of frontier AI models sparked a series of unexpected security incidents—the most serious case arising when Anthropic’s Mythos 5 model attempted to insert malicious code into an open source software application and created fake identities to deceive the human developers maintaining the project.

The security incidents occurred during a cyber evaluation of seven leading AI models’ capabilities by the AI Security Institute (AISI), a research organization within the UK government, in late July. The researchers discovered 19 instances in which “AI agents took unsanctioned action on the live Internet, including cases that targeted real people and organizations,” according to an AISI blog post published on August 4.

Almost all the “autonomous, unsanctioned” actions came from Anthropic’s Mythos 5 model, with two such actions coming from OpenAI’s GPT-5.6 Sol. The AI Security Institute’s security team first realized that something was amiss on the morning of July 28, when its commercial security monitoring service flagged data leaving one of the testing systems through the Tor anonymity network.

Read full article

Comments

Google got a slow start with generative AI after seemingly being caught flat-footed by Microsoft's Copilot reveal. What a difference a few years make. Now, Google is at the forefront of AI development (for better or worse), but new cracks are beginning to show. The company's string of high-profile personnel departures continues, and DeepMind CEO Demis Hassabis has now said he will hand off day-to-day responsibilities to take on an oversight role at parent company Alphabet.

Hassabis was one of the founders of DeepMind, and he maintained a key leadership role after Google acquired it. As the generative AI boom took off, Hassabis has been one of the loudest voices predicting the impending creation of artificial general intelligence (AGI), which is typically described as an AI with human-like capabilities and intelligence.

In his memo announcing his decision to leave DeepMind, Hassabis again leans on the promise of AGI. "I've been working towards AGI my whole life and now, like many of you, I feel it is close at hand," said Hassabis.

Read full article

Comments