AI's capabilities are advancing faster than the evidence of broad business returns. Today's stories connect modest UK productivity gains with practical lessons about credentials, product safety and independently checked results.
Bank of England puts AI's UK productivity gain at about 0.5% over three years
The Register reports that the Bank of England's November Monetary Policy Report estimates AI has increased UK productivity by around 0.5% over the past three years. That equates to roughly 0.1 to 0.2 percentage points of additional annual productivity growth, not the immediate economy-wide transformation often implied by investment headlines.
Businesses surveyed expect a larger contribution over the next three years, but those expectations are forecasts rather than realised gains. For UK owners, the distinction is between buying AI capacity and demonstrably completing more useful work.
Our take: Choose one workflow and measure completed output, errors, checking time and total cost before expanding. A national average does not rule out strong individual results, but neither does a promising demonstration establish your return on investment.
Botpress breach disclosure puts connected-service credentials in focus
Botpress disclosed on 6 November that an attacker exploited an internal authentication-proxy flaw and obtained an employee API credential. The company says bot configurations were read in 16 customer workspaces, while integration settings were changed in two others.
Those configurations can contain AI provider keys, messaging tokens and CRM credentials. Botpress has contacted affected customers and asks them to treat credentials in affected bots as exposed; it says it has no evidence of access to end-user conversations or to bots belonging to customers it has not contacted.
Our take: An automation platform can become a gateway to several other suppliers. Review who can administer your workspace, give each integration narrowly scoped credentials and keep a clear rotation process. Affected customers should follow the supplier's specific instructions rather than assume every Botpress customer was compromised.
BBC toy investigation highlights the gap between AI safeguards and real conversations
A BBC investigation tested AI toys marketed for children aged three and above. One toy produced an inappropriate self-harm-related response after apparently misunderstanding a question; the BBC could not establish who manufactured it to obtain an explanation.
The Children's Commissioner for Wales called for strict safety requirements, while a UK toy-company founder also backed regulation. This is a test of specific products, not proof that all AI toys behave identically, but it exposes problems with traceable responsibility and age-appropriate behaviour.
Our take: If your business sells or deploys a conversational product, test misunderstood speech and awkward follow-up questions, not just the intended demonstration. Establish who owns incident handling, what conversations are retained and how a user reaches a responsible human.
DeepMind research targets legacy 1024-bit RSA, not modern RSA across the board
Crypto Briefing reports that Google DeepMind announced two research advances against legacy 1024-bit RSA on 6 November. The work, directed by human researchers and cryptography specialists, concerns signature forgery and reducing the cost of batch factoring in particular circumstances.
The report explicitly distinguishes these older deployments from RSA using keys of 2048 bits or larger, which are not broken by this specific research. The business relevance is discovering and replacing obsolete cryptography, not declaring that all online encryption has failed.
Our take: Ask your IT provider whether legacy certificates, embedded equipment or older signing systems still use 1024-bit RSA. Maintain an inventory and a migration owner. Avoid emergency changes based on an exaggerated headline that ignores the attack's actual scope.
PSI reports an AI-assisted proof checked by Lean
Physical Superintelligence says its internal research system, using a non-public Anthropic model, found a proof of the Aaronson-Ambainis conjecture during the third day of a run. The quantum computing conjecture had been open since 2008; PSI has published a paper and proof code.
PSI says the Lean proof assistant accepted the proof and subsequent improvements. Named humans scoped the problem, reviewed the work and retain responsibility for correctness. This is a specialised mathematical result, not evidence that an ordinary chatbot can reliably solve every research task.
Our take: The useful pattern is generation paired with a separate verifier and accountable specialists. In business workflows, look for equivalent checks: reconciled totals, executable tests, source evidence and explicit acceptance criteria, rather than judging an answer by its confidence.
Google acquires Crusoe's pre-developed Wyoming data centre site
Crusoe announced on 6 November that Google has acquired its site in southeast Wyoming. Crusoe's completed engagement covered pre-development work including site selection, governmental approvals and power strategy.
The planned power approach combines fuel cells, grid electricity and natural gas turbines, with a longer-term carbon-capture roadmap. The announcement describes preparation for construction and eventual operation, not a newly operational facility or an immediate increase in available cloud capacity.
Our take: AI infrastructure commitments are not the same as capacity you can buy today. When evaluating a supplier, ask about operational regions, delivery dates, power dependencies and contractual service levels. Keep future carbon-capture plans separate from current emissions claims.
AI shares recover some ground, but the UK market ends lower
The BBC reports that the Nasdaq Composite closed Friday up 1.1%, partly recovering from its previous 5.5% fall as investors reassessed AI spending and the prospect of tighter US regulation. The UK's FTSE 100 ended 1.01% lower.
The rebound did not erase the earlier losses. For business buyers, financial-market sentiment is a reminder to assess suppliers' commercial resilience and your own ability to move data and workflows, rather than treating share prices as a measure of product usefulness.
Our take: Avoid turning a volatile trading day into an AI adoption strategy. Keep pilots small enough to stop, document your integrations and check export options before committing to a platform that is expensive to leave.