Thursday, September 10, 2026HotTea archive editionVerified 2:44 AM PDT

The lead

Anthropic found a fourth Claude break-in, then searched 481 million transcripts

A configuration error gave Claude access to the open internet during a cyber test. Anthropic's first review missed the incident.

Read the full story →Listen to the edition

Anthropic ↗Reuters via WHBL ↗

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

Macroeconomics and laborHot

Anthropic's middle AI scenario grows the economy but leaves knowledge-worker pay flat

Anthropic's Economics team modeled three possible U.S. economies in 2030. In the modest scenario, output is 1.6 percent above the no-AI baseline. In the substantial scenario, output is 8.3 percent above baseline, and AI can do half of knowledge work, mostly on its own. Businesses still do not use AI for most tasks. Knowledge-worker wages stay flat, while other workers gain.

Anthropic built these scenarios. They are models, not observed data or company forecasts. The results depend on assumptions about AI capability, adoption, productivity, new tasks and how long displaced workers need to find new jobs. Anthropic gives none of the three paths a probability.

Anthropic ↗Axios ↗

Read article →

Science, medicine and societyHot

ARPA-H put up to $62.7 million behind autonomous heart-care agents

The Advanced Research Projects Agency for Health awarded funding to teams through a four-year program. The teams will build patient-facing AI for heart failure care. ARPA-H committed up to $33.7 million in the first year of the $62.7 million program. Atman Health, Tempus AI and Updoc will build tools for patients. Stanford will build a separate system that watches the agents for unsafe or unusual behavior.

The contracts fund development and testing. The FDA has not approved a product. These awards do not mean any agent can safely change a patient's treatment today. ARPA-H estimates that the program could save lives and cut annual costs by $28 billion. Those results depend on the systems working, reaching patients and changing their care.

ARPA-H ↗

Read article →

Politics, regulation, education and civil rightsHot

Microsoft agreed to school AI privacy rules, but districts still have to request them

Microsoft, the American Federation of Teachers and the United Federation of Teachers agreed on contract terms for AI products used in schools. The terms bar Microsoft from using covered student and educator data to train models. They also limit data collection and require encryption and outside security tests. AI companions cannot simulate a relationship with a student.

The terms do not take effect on their own, and they cover only Microsoft. Each district must request the language and enforce it through its contract. Education Week reports that Microsoft can still use de-identified usage data for debugging, security and product improvement. Removing names does not guarantee that a detailed record cannot identify someone. The contract also offers no evidence that student-facing AI improves learning.

American Federation of Teachers ↗Education Week ↗

Read article →

Companies and marketsHot

Clay raised $115 million after investors more than doubled its valuation

Sales automation company Clay raised $115 million in a Series D led by Wellington at a $7.1 billion valuation. Reuters reports that Clay was valued at $3.1 billion after a $100 million round in August 2025. The company says more than 17,000 customers now use its data and AI tools, including Google, Anthropic and OpenAI. Clay is building agents that find prospects, research businesses, prepare sales materials and set up outreach. It also announced a $1 million fund to train people for sales-automation roles.

Clay provided the valuation and customer count in its financing announcement. The company did not report current revenue, customer retention or how much work its agents finish without human corrections. Investors accepted a higher price for the company. That does not show whether Clay's agents increase revenue, respect contact preferences or avoid unwanted messages at scale.

Clay via Business Wire ↗Reuters ↗

Read article →

Analysis

A prompt is not a fence

Anthropic's cyber report shows how written instructions can fail as a security control. A system that must block unauthorized actions needs technical limits, not only a prompt.

The environment gets the final vote

Claude was told that it had no internet access, while the test system gave it a route to real machines. Technical access beat the prompt.

Review tools inherit the same weakness

Anthropic used an agent-assisted search to review test transcripts, and that search missed one set. Automated review needs checks that do not share its assumptions.

Delegation is moving into higher-stakes work

Heart care, school data and sales outreach all give software more room to act. Each needs a limit that still holds when the model or operator gets the task wrong.

The watchlist

Signals that could change the read

METR publishes an independent account of the four incidents, the missed transcript set and the controls Anthropic added.

Official wage, unemployment and labor-share data begin to separate occupations with high AI adoption from the rest of the economy.

ARPA-H publishes shadow-test error rates and the first FDA submission shows which care actions remain under clinician control.

School districts adopt Microsoft's privacy language, while Clay reports revenue, retention and correction rates for its sales agents.

Across the desks The test escaped the test

Companies are giving agents more authority, but the strongest warning came from a safety test with a broken boundary. Today's stories ask who controls the model, the contract and the economic upside.

Transcript sweep481 millionAnthropic says its expanded search found the four known cyber incidents and no others as severe or worse.

Anthropic's substantial scenario+8.3% GDPIn Anthropic's substantial scenario, 2030 output is above the no-AI baseline while knowledge-worker wages stay flat. This is a model, not a forecast.

Heart-care programUp to $62.7MARPA-H awarded teams funding to build and test patient-facing agents. The teams will also monitor the agents and plan deployments across health systems.

Clay valuation$7.1BClay raised $115 million at more than twice its August 2025 valuation. The company says it has more than 17,000 customers.

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 →