AI Pulse · Frontier Labs Watch

Anthropic has published a 154-page threat-intelligence disclosure describing state-linked misuse of Claude. The detail matters less as a one-off security story and more as evidence in the wider argument about who should oversee powerful AI systems. You can see the initial reporting in this AFP update.

For business owners, this is a reminder that AI risk is not limited to someone on your team pasting confidential information into a chatbot. As AI tools become more capable, misuse can come through fraud, impersonation, automated research, or people using systems in ways their creators did not intend. The debate is now moving beyond whether companies should publish safety policies. It is increasingly about whether independent parties should be able to inspect, test, and challenge the safeguards.

That doesn’t mean every business needs a formal AI audit tomorrow. It does mean you need clearer ownership. Know which AI tools are in use, what business data they can access, who approves new deployments, and what happens when something looks wrong. This is the kind of thing we build into an AI command centre, giving leaders a practical view of where AI is being used and where controls need attention.

The timing is important. Arguments over how quickly frontier AI should advance are becoming arguments over what proof of safety is required before the next capability is released. Public disclosures like this add pressure for oversight that sits outside the labs themselves.