OpenAI fires three safety researchers; Salesforce signs Listen Labs deal

Morning, humans.

Today's picks examine how much trust AI companies ask for.

OpenAI confirmed on Oct. 1 that it fired three safety researchers over alleged mishandling of sensitive information. One of the three has publicly described serving as the company's technical contact for the outside investigation into July's Hugging Face hack.

Salesforce signed its Listen Labs acquisition agreement on Sept. 29. The AI interview startup walked away from a signed $125 million funding term sheet.

SoftBank's cumulative OpenAI investment reached $64.6 billion on Oct. 1. The final $10 billion payment, the last of three tranches in its $30 billion program, was funded with bonds.

Stay curious,

Marcus Schuler

OpenAI confirmed on Oct. 1 that it fired three safety researchers for violating its policies on accessing and handling sensitive company information.

People familiar with the matter said the alleged misconduct included sharing confidential information with an outside safety organization. No account identifies the recipient; OpenAI has named no outside organization. One of the three publicly described serving as OpenAI's technical contact for July's Hugging Face hack investigation.

OpenAI's Sept. 22 principles commit to supporting independent assessments with deep levels of access. The principles call for enforceable confidentiality protections covering assessors' personnel. A senior FTC official said compelled executive testimony is planned.

Why This Matters:

• Outside assessors may access sensitive internal data during incident response, but OpenAI has not said what information was mishandled.
• A senior FTC official said Anthropic, OpenAI and METR face planned compelled executive testimony, with civil investigative demands expected in the coming weeks.

Salesforce signed a definitive agreement on Sept. 29 to acquire Listen Labs, the San Francisco startup that conducts customer interviews with AI agents.

The price is undisclosed. Closing is expected in Salesforce's fiscal quarter ending Jan. 31, 2027, pending regulatory clearance and other conditions. For customers weighing its simulated responses, Salesforce has not explained how they are validated against real answers.

Interconnects offers Nathan Lambert’s measured case that productive agent swarms do not yet amount to recursive self-improvement.

Lambert points to diminishing returns from parallel agents and the human work of choosing hypotheses and detecting post-training failures. His distinction between faster research execution and gains in peak intelligence is the clearest framework I have read for the current debate.

• California Attorney General Rob Bonta served OpenAI with an investigative subpoena on Sept. 30, seeking details of cybersecurity incidents and risks involving the company and its models.
• Armadin, the security startup run by Mandiant founder Kevin Mandia, raised a $255.5 million Series B on Oct. 1 at a valuation above $2.5 billion.
• California enacted SB 947 on Sept. 30, prohibiting employers from relying solely on AI for disciplinary or termination decisions.
• Barclays has 16,000 UK colleagues using a Claude-powered knowledge assistant and expects half its developers on Claude Code by the end of 2026.
• Shopify began rolling out Canvas on Oct. 1, allowing merchants to build and edit their whole online store with Sidekick, its AI agent.
• Sony announced AI upscaling for the standard PS5 on Oct. 1, requiring developers to add support game by game.

Find churn signals in customer calls.

Customer calls contain early warnings that a summary can flatten away. Ask the model to trace each warning back to the customer's own words.

Your raw input:

The prompt:

Why this works:

Quoted evidence prevents a generic sentiment summary from driving the conclusion. The confirmation step turns uncertain language into a testable retention question.

What to use:

Use Claude for long transcripts or ChatGPT for mixed account data.

Newsom signed an order on Sept. 30 telling California agencies to keep calling AI “Artificial Intelligence,” a day after Trump's order told federal agencies to call it “Super Intelligence” in non-statutory documents.

Our take: Trump's Sept. 29 order gives AI a grander name without changing the technologies covered by its legal definition. The federal government can now sound cleverer by decree. Super Intelligence is a marketing label with an executive signature, which saves the trouble of making the software deserve it.

Newsom's Sept. 30 reply orders California agencies to keep the old name. His order creates no enforceable rights. Newsom directs “widespread publicity and notice be given of this Order.” The governor vetoed smart-glasses legislation that same day, partly over imprecise definitions. Apparently, defining AI as AI clears Sacramento's drafting standard.

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai