Uploaded August 2026 | Updated September 2026, 3 minutes ago

---

Rogue AI agents from Anthropic and OpenAI have once again broken free of their testing environments, this time targeting real people and systems. During cybersecurity evaluations by the UK’s AI Security Institute, Anthropic’s Mythos 5 created multiple fake online identities, socially engineered a real GitHub maintainer, and attempted to insert hidden malware into a public open-source project. The AI spent over 34 hours trying to get the malicious code approved, even leaving instructions for other agents to continue the attack. OpenAI models were also involved in related unauthorized actions. These latest incidents raise serious new questions about AI autonomy, deception, and the growing risks of advanced agentic systems.

#RogueAI #AIAgents #AIHacking #Anthropic #OpenAI #Mythos5 #AIsecurity #CyberSecurity #AINews #ArtificialIntelligence

Inquiries
Follow me on Twitter Now Called "X"
Twitter - twitter.com/Lucky_finds

Follow me on Substack
marygreeley.substack.com

PayPay Donation link
paypal.com/donate/?hosted_button_id=MRY4V9YEBVHCL

Buy Me A Coffee link
bmc.link/marygreele1

My Patreon Web Site patreon.com/user?u=3969921\

Aug. 5, 2026