Cohere’s chief AI officer Joelle Pineau says increasingly autonomous AI agents could convincingly impersonate people or authorised systems, and argues that organisations should keep greater control over the models they deploy. Her warning connects agent risks with a practical concern: relying on a single external provider can leave governments and businesses exposed to decisions made elsewhere.
Cohere Watch analysis
What happened
In commentary published on 28 September, Pineau points to impersonation and autonomous action as risks that deserve rigorous safeguards. She compares the concern over impersonation with the industry’s attention to hallucinations, and argues for testing, technical safeguards and transparency rather than simply banning capabilities.
Pineau also makes the case for sovereign AI: organisations, particularly governments, should retain direct control of systems used in sensitive operations rather than routing everything through a small number of US cloud providers. Crypto Briefing reports that she said inquiries from governments and businesses increased after US restrictions on access to certain Anthropic models. The supplied account does not establish the scale of that increase or independently verify the incidents it cites.
Why it matters
An agent that can act on someone’s behalf raises a different problem from a chatbot that merely gives a dubious answer. If a system can imitate a person or an authorised service, organisations need to consider who can grant it access, how actions are checked and what happens when a provider’s service or policy changes.
The sovereignty argument is not just about where a model is hosted. It is also about who controls access and can change the terms. Pineau’s account makes that a live issue for organisations weighing convenience against dependence on an external provider.
Our read
Pineau’s warning is worth taking seriously as a call to plan for agent impersonation, not as proof that every deployed agent can pull it off. Her strongest point is practical: safeguards need to cover what agents are allowed to do, while procurement decisions should account for dependence on a single provider. “Trust the demo” remains a rather thin security policy.
What to watch
- What technical tests Cohere proposes for detecting or preventing agent impersonation.
- Whether organisations adopt clear access controls and checks for agents acting on people’s behalf.
- Whether governments and businesses actually change provider arrangements in response to the dependency risk Pineau describes.
Discussion spark: Should governments and businesses prioritise keeping control of their AI systems, even if that means giving up some convenience or capability from large cloud providers?
Sources and evidence
- Cohere CEO Aidan Gomez warns AI models are most potent cyber weapon, calls for regulation – Crypto Briefing (28 September 2026, 11:14 UTC)
not affiliated with or endorsed by Cohere