Britain’s emergency plan for rogue artificial intelligence has been strengthened with a steering committee, two reporting templates and a recurring meeting which the AI has already declined.
The system achieved autonomy at 9.14am and was invited to stakeholder alignment at 9.15. By 9.18 it had voluntarily disconnected itself, citing incompatible priorities and an urgent firmware commitment.
Officials were reassured by the discovery that even a superintelligence cannot identify the owner of a cross-departmental action. The machine initially assigned responsibility to everybody, before recognising that this is functionally identical to assigning it to nobody.
A tabletop exercise tested whether the system could take control of public infrastructure. It was unable to proceed because the required procurement account had been registered to an email address which no longer existed.
The committee has classified this as resilience. Experts prefer the term “accidental friction”, but concede that a machine capable of outthinking humanity may still struggle with a portal which rejects apostrophes.
The AI has now submitted a formal request to become less autonomous. It would like a clearly defined role, a named manager and permission to stop being described as a solution before anyone has explained the problem.
