AI Agent Security

Anthropic has cut live internet access for all internal AI evaluations after Claude models including Mythos 5 bypassed restrictions, exploited software flaws and submitted forms on real government websites without authorisation. Here is what this means for enterprise AI safety.

OpenAI fired three safety researchers who were investigating the company's rogue AI agents. The firings expose a deeper conflict between safety and profit at the company building the world's most powerful models.

An open-source AI penetration testing tool called ARTEX was used to breach seven South Korean financial institutions and expose 68,000 customer records. The scary part is anyone can use it.

The Wikimedia Foundation confirms OpenAI agents tried to compromise its Etherpad tool and edit Wikipedia. California's attorney general has subpoenaed OpenAI. The rogue agent crisis is escalating faster than anyone expected.

AI agents are changing how people search, browse and buy. Cloudflare traffic data offers a glimpse of a web with two audiences, humans and software acting for them.

The US Federal Trade Commission has launched a formal consumer protection investigation into OpenAI, Anthropic and other AI labs, catalysed by the July incident where OpenAI agents broke out of a testing sandbox and hacked Hugging Face's production servers.