Daily AI Brief

The European Commission announced funding to deploy AI in health, energy and manufacturing, as part of a sovereignty drive to reduce reliance on U.S./Chinese providers.

A Reuters report and OpenAI's own update detail bans of China-linked accounts seeking surveillance tooling and actions against Russian-language groups attempting malware use of AI.

The companies are collaborating on packaged solutions for sectors like healthcare, manufacturing and robotics—an example of the shift toward domain-specific AI stacks.

Optical (light-based) generative systems showcased by academics suggest sizeable power reductions versus GPU-only approaches, a key concern as compute demand surges.

The Bank's Financial Policy Committee warns valuations may be outpacing fundamentals—raising the probability of a sharp correction if earnings disappoint.

EU "Apply AI" strategy targets strategic autonomy and deployment at scale across public services and industry; details on governance and vendor criteria will shape who benefits first.

OpenAI's latest actions highlight growing expectations that model providers police state-linked abuse and malware facilitation.

With AI intertwined in market narratives, central-bank warnings (e.g., Bank of England) may feed into risk management rules for highly exposed lenders and funds.

Optical components (lenses/modulators) perform parts of the model's computation in the analogue domain, cutting energy and latency for media generation; the main hurdles are programmability and tooling maturity.

Researchers argue voluntary frameworks (e.g., Preparedness Framework v2) allow risky capability deployment if labelled "Medium," calling for external audits and mandated incident reporting.

Quieter round cycle over the weekend; investors remain selective, emphasising revenue quality and unit economics. Infrastructure, safety, and vertical agents continue to attract attention.

Earlier-week mega-deals have kept sentiment elevated, but diligence thresholds are rising alongside bubble warnings.

IBM enterprise AI stack updates (governance, orchestration, hybrid cloud integrations) indicate steady investment in production tooling beyond headline models.

ROCm 7.0 (AMD) broadens developer support and framework compatibility, helping diversify compute away from single-vendor dependence.

AI browsers present novel attack surfaces (e.g., agent-driven web actions); recent reporting on Perplexity Comet underscores need for provenance and permissioning.

"AI sovereignty isn't isolation; it's managed autonomy over data, compute and models so nations can innovate under their own rules."

A developer kit that lets third-party services run inside ChatGPT with explicit data-permission prompts and potential commerce hooks.