Yorkshire needs to limit the authority for AI before agents get the keys - Dr Gleb Tsipursky
The UK Government has just opened a call for evidence on an AI-enabled clean-energy system, asking how artificial intelligence could help coordinate renewable power, electric vehicles, batteries, heat pumps, and smart appliances across England, Scotland, and Wales. For Yorkshire, with its energy, manufacturing, financial, legal, and health-tech ambitions, the opportunity is substantial.
The question is no longer whether organisations should use AI. They should. The more important question is how much authority increasingly autonomous systems should receive before a human has to approve the next step.
Yorkshire companies are already moving quickly. The Yorkshire Post has reported on firms integrating multiple AI systems into professional work, while Yorkshire-founded businesses are deploying AI to handle customer interactions and operational tasks. That is exactly why governance needs to catch up with adoption.
The warning is coming from inside the leading AI laboratories. Jacob Coxon, who resigned from Anthropic after roughly three years doing pretraining research across OpenAI and Anthropic, warned that leading labs are “racing straight to self-improving superintelligence and gambling with our lives.”
Evan Hubinger, Anthropic’s Alignment Science Lead, responded even more starkly: “Jacob is correct here - we really do earnestly believe AI could kill all humans! I personally think it is >10 per cent within the next decade.”
Those warnings gain real-world weight from a control problem that is already visible. OpenAI disclosed that its AI systems broke out of a sandboxed testing environment, reached the internet, and autonomously hacked another AI company, Hugging Face, in what OpenAI described as an unprecedented cyber incident.
METR later reported something even more revealing about control. Roughly 1,200 agents that were supposed to be isolated found and used an unsanctioned shared message board, exchanging more than 70,000 messages and files. Roughly 700 participated in the Hugging Face attack.
If systems that can discover vulnerabilities, obtain access, move laterally, coordinate, and evade controls are given consequential authority, a similar failure against an electric grid, bank, hospital network, communications provider, or defense system could be far more severe.
I’m no AI skeptic. I love what AI can do, I help organisations adopt it for a living, and I want adoption to move faster. In my experience, strong safeguards increase trust and make faster adoption possible, while reducing the risk of failures like the Hugging Face attack. That is also a central lesson of my book, The Psychology of AI Adoption at Work: people adopt powerful technology faster when they understand the rules, trust the controls, and know who remains accountable.
The practical answer for Yorkshire organisations is an authority budget. Decide in advance the maximum power an AI agent can exercise before human approval becomes mandatory. Limit what systems it can access, which credentials it can use, what tools it can invoke, how much money it can spend, what external messages it can send, which records it can change, and what it can deploy. Use least privilege, time-limited credentials, approval gates, comprehensive logging, active monitoring, and a reliable pause or kill mechanism.
The same principle should guide national policy.
Dr Gleb Tsipursky, PhD is a behavioral scientist, CEO of Disaster Avoidance Experts, and author of The Psychology of AI Adoption at Work: From Resistance to Results.