Optimus Consulting
The Week in AI
Edition 08
UK / Service SMEs
This week AI stopped being consequence-free.
Courts, regulators and a government stepped in within the same five days. A US court made Google liable for what its AI told people. The US government pulled Anthropic's newest model. The lesson for everyone else is simple. AI output now carries weight, and someone has to own it.
Top stories
5 to knowA court ruled Google liable for what its AI said.
A US court found Google legally responsible for false statements produced by its AI Overviews. It is the clearest sign yet that the business publishing an AI answer, not the model, owns the consequences when it is wrong.
Optimus angle
Before any AI-written message reaches a customer, add one human check. Name who signs off on AI output for each client-facing process, and make that sign-off the final step before send.
The US government pulled Anthropic's newest model.
Anthropic switched off its most powerful new model days after launch following a US Commerce order tied to safety and national security concerns. A frontier tool can vanish overnight, by order, not by choice.
Optimus angle
Do not build a core process on a single model you cannot replace. Note which tasks would stall if your main AI tool went dark tomorrow, and keep a second option you know how to switch to.
KPMG pulled a report after AI hallucinations.
KPMG withdrew a published report after AI-generated errors undermined its findings. A trusted firm, a checked process, and the mistakes still reached print. Reliability is not a setting you switch on.
Optimus angle
Treat AI drafts as a first pass, never a final one. For anything that carries your name, verify every figure and quote against the source before it leaves the building.
OpenAI redesigns ChatGPT while regulators circle.
State attorneys general have opened investigations into OpenAI over its ad and health-data practices, even as it reshapes ChatGPT into a broader superapp and edges toward an IPO. The tool your team uses daily is changing under them.
Optimus angle
When a tool you rely on adds features, do not adopt all of them. Decide which new capabilities help a real task, leave the rest off, and tell the team what is sanctioned.
Apple rebuilt Siri into a real assistant at WWDC.
Apple used WWDC to relaunch Siri as a more conversational, context-aware assistant with its own app. AI help is moving onto the phones your staff and customers already hold, no new software required. That lowers the bar to everyday use, and to everyday mistakes if it touches work data.
Optimus angle
Phone-based assistants will start handling work tasks whether you plan for it or not. Decide now what staff may and may not put into a personal AI assistant, especially client and claims data.
From the Insurance Desk
Sector signal for brokers, motor claims and credit hire companies.
Item 01
Cytora says Zurich proves agentic AI is past the pilot.
Cytora's chief executive says its deployment with Zurich shows agentic AI now runs real underwriting workflows, not test cases. Autonomous assistants are moving into live broker and insurer processes.
Optimus angle
If agents are entering underwriting and submissions, map where one could triage your inbound work first. Pick the highest-volume, rules-based step and test it on last month's cases before it touches live ones.
Item 02
Aviva blocked record fraud as criminals turn to AI.
Aviva reported it stopped record levels of claims fraud in 2025 as criminals used AI-generated images and documents. The same tools that speed up honest claims are speeding up the fake ones.
Optimus angle
Assume some evidence you receive is now AI-made. Add a second-eyes check on photos and documents in higher-value motor and credit hire claims, and log what made you suspicious so the pattern builds.
Editor's column
The Optimus Take
For two years the risk story around AI was about what it might do. This week it became about what it has done, and who answers for it. A court, a government and a Big Four firm all drew the same line within days. The output is the liability, and the name on the output is the one that carries it.
For a UK service SME that is not a reason to slow down. It is a reason to add one step. The firms that win with AI this year will not be the ones using the most tools. They will be the ones who can point to a named human behind every answer that reaches a client. Speed is easy. Accountable speed is the advantage.
- StrategyTreat every AI output that leaves your firm as your firm's own words. Decide which client-facing processes need a named sign-off before they send.
- OperationsVerify AI-drafted figures, quotes and legal points against source before publication. Draft is a first pass, never the final one.
- ServiceTell clients where AI helps and where a person owns the answer. Accountability is a service standard, not a disclaimer.
“The model makes the mistake. You still make the apology.”