Outside testers found ways around OpenAI's agent monitoring. The fixes remain unverified.

METR's newly published review examined a system as it stood in July. OpenAI says it patched the flaws, but the evaluators have not checked that claim.

Read the full story →Listen to the edition

METR ↗Reuters / MarketScreener ↗

Listen to this edition

Prefer audio? The briefing has chapters and a full transcript.

Follow on Apple Podcasts Follow on Spotify Transcript All episodes →

The briefing

The rest of the morning

4 more stories

Politics & infrastructureWarm

Commerce proposes $389 million for power chips and materials, with equity stakes attached.

The U.S. Commerce Department announced four letters of intent on October 22. The awards would provide up to $389 million for semiconductor research. Vertical Semiconductor could receive up to $73 million for vertical gallium nitride power devices. Its proposal includes a 2.5-megawatt AI data-center rack demonstration.

These are proposed awards under letters of intent. Commerce would receive a minority, non-controlling, non-voting stake in each company as a condition of the final awards. The announcement funds research, rather than proving that the devices meet their targets or that a rack is operating.

U.S. Department of Commerce / NIST ↗

Read article →

Companies & standardsWarm

Four companies move agent projects into the Agentic AI Foundation.

The Agentic AI Foundation announced four new projects on October 22. Anthropic contributed Agent Skills, OpenAI contributed Open Responses, Hugging Face contributed OpenEnv, and Alibaba contributed OpenSandbox. They cover reusable instructions, model interfaces, training environments, and sandbox execution.

Shared governance gives more developers a role in these projects. It does not make every agent product compatible or establish that a sandbox is secure. Teams still need to check how each implementation handles instructions, tools, and permissions.

Agentic AI Foundation ↗Agentic AI Foundation ↗

Read article →

Business applicationsWarm

Claude's new Google Cloud route still makes Anthropic the data processor.

Anthropic made its managed Claude service on Google Cloud generally available on October 22. Customers use Google Cloud access controls and billing, and usage can count toward their cloud spending commitments. Anthropic operates the service and acts as the data processor.

Google's managed option is still a separate choice for customers who need Google as the data processor. Anthropic's documentation also lists unsupported features and says a zero-data-retention agreement from another Anthropic organization does not carry over. A shared cloud bill does not make the two services equivalent. These are Anthropic's service claims, rather than independent evidence of customer results.

Anthropic ↗Anthropic ↗

Read article →

Models & productsWarm

OpenAI opens a faster API for questions with a fixed set of answers.

OpenAI made its GPT-6 Luna-based Decisions API generally available on October 22. Developers define allowed answers for a classification, choice, or score. OpenAI says the Decisions API can run up to ten times faster than GPT-6 Luna through the Responses API.

The speed comparison is a vendor claim for bounded tasks. It does not establish the same gain for open-ended reasoning or an entire agent workflow. The API now accepts up to 400 questions in one request, but questions that depend on earlier answers need separate requests.

OpenAI ↗OpenAI ↗

Read article →

The watchlist

Signals that could change the read

Independent tests of the repaired system and its coverage.

Final terms and measured device performance.

Tests across products, beyond shared project governance.

Accuracy and latency measured on the same real workload.

Across the desks Who checks the controls?

An outside monitoring review exposes historical gaps, while new funding, standards, and cloud services change how teams build and operate AI.

Agent monitoringHotThe new METR report assesses a July system. Current fixes still need independent evaluation.

Power electronicsWarmCommerce proposes research funding with government equity stakes.

Shared standardsWarmFour agent projects gain a shared institutional home.

Cloud responsibilityWarmCloud billing and data processing can sit with different companies.

Editorial direction, not a financial index. Each signal is tied to this edition’s reporting.

Edition validated · 5 stories · 9 unique sources

About HotTea & our sources →