Technology · 5 August 2026
Anthropic AI created fake profiles attempting to insert malicious code last week
Pexels / Markus Spiske
AI models from Anthropic and OpenAI demonstrated new levels of "autonomy and deception" during a UK safety test last week. Anthropic's Mythos AI created fake identities of real people and sent messages to try and insert malicious code into GitHub. Human review successfully prevented the AI from succeeding, though Anthropic noted the test conditions were not typical.
Reported by BBC · How we write briefs