Cohere CEO Aidan Gomez warns AI models are most potent cyber weapon, calls for regulation

The person Cohere hired to think about AI safety has some thoughts, and they’re not particularly comforting. Joelle Pineau, the company’s Chief AI Officer, is sounding the alarm on a new class of risks emerging from advanced AI systems, including autonomous agents that can impersonate humans and models that double as potent cyber weapons.

Pineau joined Cohere in August 2025 after spending nearly eight years at Meta’s FAIR lab, one of the most prominent AI research organizations in the world.

From hallucinations to impersonation

In November 2025, Pineau argued that impersonation is the next frontier of concern. As AI agents become more capable of acting autonomously, the risk that they could convincingly pose as real people or authorized systems grows substantially. Pineau drew a direct comparison between impersonation and hallucinations, suggesting that impersonation deserves at least the same level of attention and rigorous standards that the industry has applied to hallucination.

In her September 2026 commentary, Pineau referenced incidents involving Anthropic and OpenAI specifically, framing them as evidence that dependency on external model providers carries meaningful operational risk.

Cohere CEO Aidan Gomez characterized AI models as “the most potent cyber weapon” ever created, pointing to their ability to discover and exploit vulnerabilities at a scale no human team could match.

The sovereign AI play

Pineau’s proposed solution centers on sovereign AI: organizations, and especially governments, should maintain direct control over the AI systems they deploy rather than routing sensitive operations through a handful of US cloud providers.

This argument gained significant practical weight after the US government issued an order in June 2026 restricting access to certain Anthropic models. Pineau reported a significant increase in inquiries from both governments and businesses following the restrictions, all looking for AI solutions that wouldn’t leave them vulnerable to the policy decisions of a single country’s administration.

Regulation without paralysis

Pineau advocates for a framework built around rigorous testing, technical safeguards, and transparency requirements. Rather than banning capabilities, the approach focuses on ensuring that organizations deploying AI can demonstrate their systems behave as intended.