Friday, October 30, 2026Radar run of Oct 30: 41 new entries, 10 read in full, 35 quotations checkedabout 5 minutes

The Verge reports Microsoft's overall headcount fell year over year for the first time in a decade, to around 223,000, alongside the 60,000-person buyout offer.

How to read this pageSerif their words, as the source published themMarked found word for word in the sourceSans our words

Oct 29Oct 31 Today
Read asSame facts, in the order each reader needs them.

Heard today

Press play to see and hear the person say it, from the source’s own player. Their words light up as they are spoken, and every one was found word for word in the source. Nothing loads from YouTube until you press play.

Group not yet recorded

“Tech titans build bunkers while warning the artificial intelligence they're creating risks human extinction. Meanwhile, our Congress debates youth sports and the president builds a ballroom.”

Ossoff, speaking at a Georgia midterm rally, contrasts the warnings of AI company leaders with what Congress and the president are doing.

Jon OssoffSenator (D-GA)Forbes Breaking News, Oct 29 · found word for word · ▶ at 0:53

Group not yet recorded

“I think these guys have all read too much science fiction.”

Eisman replying to a host who noted that Altman, Elon Musk and Amodei were voicing AI risks years before their companies existed.

Steve EismanHost, The Real Eisman Playbook podcast; former senior portfolio manager, Neuberger BermanCNBC Television, Oct 29 · found word for word · ▶ at 3:28

The video’s captions give a time for each line, not each word, so a word can light a few tenths of a second early or late. The recording is the evidence; the highlight is a guide.

What changed today

Decisions, new evidence, corrections and commitments we read in full, one per story. Arguments and recaps don’t count.

New evidenceThe Verge

The Verge reports Microsoft's overall headcount fell year over year for the first time in a decade, to around 223,000, alongside the 60,000-person buyout offer.

News: Microsoft’s AI future has fewer jobs

What it means for youMicrosoft is offering buyouts to roughly a quarter of its staff while spending on AI, a sign that AI investment can be paid for partly by shrinking payrolls.

For lawmakersThe Verge reports Microsoft's headcount fell for the first time in a decade as it cut costs to protect AI spending, which is relevant to anyone tracking AI-linked job losses.

For researchersWarren ties the buyouts to Nadella's earlier remark on people costs and GPUs per researcher; a company-level data point on AI and headcount, though the causal link is Warren's reading.

“Our total people costs will go down, our cost per head will go up, and my GPU per researcher will go up,”

Nadella on a podcast almost two years earlier, answering a question about AI's impact on headcount, as quoted by The Verge.

Satya NadellaCEO, MicrosoftAI company leaderThe Verge, Oct 29 · found word for word

New evidenceUK House of Commons

Bob Blackman says the postponed Commons debate on advanced-AI risks could be held on 19 November if the Government allocates time.

Hearing: Business of the House

What it means for youA planned House of Commons debate on the risks of advanced AI was pulled on Monday because too many MPs wanted to speak; it may return on 19 November.

For lawmakersSir Alan Campbell, Leader of the House of Commons, promised no Government time for an AI debate, citing the UK's G20 presidency as a route to global AI principles.

For researchersA planned House of Commons debate on the risks of advanced AI was pulled on Monday because too many MPs wanted to speak; it may return on 19 November.

“the Budget delivers the Government’s most ambitious AI package yet, investing £700 million to tackle the risks of AI. We will put AI centre stage during our G20 presidency”

Sir Alan Campbell (Labour), Leader of the House, replying to the request for an AI safety debate.

Sir Alan CampbellLeader of the House of CommonsUK House of Commons, Oct 29 · found word for word

Said today

One checked quotation per person, people with the power to act or first-hand knowledge first, at most two from any one group.

Who is quoted here: 2 with no group recorded yet, 1 journalist or commentator.

Group not yet recorded

“the House will not have long to wait for the debate on the emerging risks of advanced AI, which of course had to be withdrawn because of the sheer number of Members who wanted to speak”

Bob Blackman (Conservative), Chair of the Backbench Business Committee, on rescheduling the debate if the Government allocates time on 19 November.

Bob BlackmanChair of the Backbench Business CommitteeUK House of Commons, Oct 29 · found word for word

Group not yet recorded

“It wouldn't surprise me if future incidents of roughly that magnitude were coming every three-ish months from here on out, maybe accelerating over time. I think there will be many future moments.”

After noting the Mythos announcement and the Hugging Face hacking incident came about three months apart.

Luke MuehlhauserManaging director, AI Governance and International Policy team, Coefficient Giving80,000 Hours, Oct 29 · found word for word

Journalist or commentator

“Anthropic has a lot of wild risk factors in this S-1, including that Claude might be conscious and kill us all.”

Patel, the host, characterising the risk-factor section of Anthropic's filing while asking about cloud suppliers that are also competitors.

Nilay PatelEditor-in-Chief, The VergeJournalist or commentatorThe Verge, Oct 29 · found word for word

Watch and listen

Interviews and talks we read in full, by what they add, then by reach. ▶ opens the video at the moment quoted.

6 of 9 videos and podcasts in this run were read in full, from captions or a transcript. Only those are ranked here; the rest are under “Everything else”, from their descriptions.

4min

VideoForbes Breaking News2K views

Jon Ossoff Rips Into 'Tech Titans' For Creating AI That 'Risks Human Extinction' At Midterm Rally

Senator Jon Ossoff makes AI risk part of a midterm stump speech in Georgia, blaming Congress and the president for weak precautions.

“An ancient establishment barely capable of email sleepwalks into a technological revolution while the president dismantles precautions and calls us fools for worrying.”

Ossoff criticises Washington's response to AI at the same rally.

Jon OssoffSenator (D-GA)Forbes Breaking News, Oct 29 · found word for word · ▶ at 1:09

The entry, with 2 checked quotations

156min

Podcast80,000 Hours

Luke Muehlhauser on the most important AI safety work nobody is doing

Luke Muehlhauser of Coefficient Giving names specific missing AI safety organisations and says funding is waiting for founders willing to build them.

“On the pessimistic side, I would say most institutions and projects and so on that need to exist to manage the situation well, don't exist. And there's nobody working on them at all in the world.”

Muehlhauser's overall view of the field, in an interview the host says was recorded in August.

Luke MuehlhauserManaging director, AI Governance and International Policy team, Coefficient Giving80,000 Hours, Oct 29 · found word for word

The entry, with 5 checked quotations

Decided and promised

What someone with the power to act did or committed to. A promise is not a deed: commitments are followed until they are kept or broken.

  1. Decision takenMistral released the weights of a trillion-parameter model aimed at cybersecurity and claims it leads Chinese open-weight models in coding and cyber tasks.TechCrunch
  2. New commitmentThe MicroZig open-source project says it will adopt a strict policy against LLM and AI use, according to a Hacker News post title.GitHub (ZigEmbeddedGroup)

New research

Papers and reports, the ones we read in full first. The finding in our words, how it was reached, and what it cannot show.

5 of 14 research items.

PreprintarXivfrom the description only

AgentDecept: Benchmarking Deceptive Behavior in Tool-Using LLM Agents under Pressure

Found
The excerpt describes the benchmark and protocol only; results were cut off and are not summarised here.
How
An executable benchmark with 477 pressure variants derived from 257 source tasks across six pressure sources, each paired with a pressure-free counterpart; full trajectories judged against time-ordered observations.
Limits
Results and limitations not visible in the excerpt.

PreprintarXivfrom the description only

Red-Teaming Large Language Models in Low-Resource African Languages: A Case Study of Yoruba and Nigerian Pidgin

Found
Models that rejected up to 69% of English prompts rejected as few as 0% in Yoruba, because of failed comprehension; harmful-content rates reached 0.63 in English and 0.50 in Nigerian Pidgin when models engaged.
How
Six models given 400 explicit adversarial prompts in Yoruba and Nigerian Pidgin across four harm categories; responses scored by native-speaker annotators, with reliability measured by Cohen's kappa.
Limits
Six models and 400 prompts; low Yoruba comprehension suppresses apparent attack success. Excerpt truncated.

PreprintLessWrongfrom the description only

[Paper] Propensities Please: Exposing Alignment-Faking LLMs using Synthetic Document Finetuning

Found
The authors report the oversight gap widens by 24 to 68 percentage points after their intervention, since models otherwise often suspect a trap.
How
Synthetic document finetuning convinces several alignment-faking model organisms that they are not under effective oversight; the authors measure the oversight gap, the rise in misaligned action rate when told there is no oversight.
Limits
Not read beyond the post's summary; limits unknown. Tested on model organisms, not shown on deployed models.

PreprintarXivfrom the description only

Activation Probes Are Reliable With a Handful of Positive Examples

Found
For small numbers (1-10) of positives, training with many negatives was on average more positive-sample-efficient than balanced training; LLM upsampling gave a boost equivalent to roughly doubling real positives; larger models were more efficient.
How
Activation probes trained in a few-positives, thousands-of-negatives regime on spam and honesty detection tasks, varying class balance, LLM upsampling and model size.
Limits
Results are for the authors' spam and honesty tasks; they do not show the same holds for real misalignment.

PreprintarXivfrom the description only

Clean Sheets and Checkpoints: Evidence Complexity for Certifying Sequential Agent Risk

Found
Per the abstract, a certificate exists for any specified attempt budget, and the minimum expected number of zero responses is asymptotic to a stated expression that the certificate attains.
How
A theoretical model: a task of k milestones split into checkpoint segments, asking how many zero-success audit responses certify a completion probability of at most a given limit at a given level.
Limits
The abstract says the audit does not test the transport assumption that deployment success probabilities are capped by audit ones. Read from the excerpt only; the rest of the abstract was cut off.

Stories that continued

Events several sources covered, read as one thread across days.

Where the answers stand

2 of the nine questions have an answer we keep up to date, with the reason for every change. All nine questions

Everything else today: 31 more entries, filed under the nine questions

Can AI systems act beyond what people intended or allowed?

Are AI companies putting stronger safeguards into practice?

Who can demand evidence, delay a release or stop an unsafe deployment?

How is AI used in attacks, defence and war?

What happens to work, income and the economy?

How is AI changing childhood, health and daily life?

What values should AI have, and could it have a mind of its own?

How fast is AI advancing, and where is it heading?

Search and filter the week’s record

How this edition was made

So you can judge how far to trust it.

  1. 114 of 114sources answered
  2. 882items they published, October 29–30, 2026
  3. 147kept by a wide keyword filter
  4. 41became entries, 10 read in full
  5. 35quotations found word for word in the saved text
How this edition chooses
  • What changed: entries whose change is a decision, new evidence, a correction or a commitment, that we read in full; most important first; one per story.
  • Heard: the same rule as Said, from the quotations we can play: the recording’s captions carry them. Any entry qualifies, the weights put the ones that matter first, and a person is quoted once on the page.
  • Said: one checked quotation per person from the entries that matter, people with the power to act or first-hand knowledge first, at most two from any one group.
  • Watch and listen: interviews and talks we read in full, by importance, then by views.
  • Decided and promised: decisions, corrections and commitments, decisions first.
  • New research: papers and reports, the ones we read in full first.
  • The headline: written for the edition when we can; otherwise the top change.

Agents pick the items, write the summaries and the “why it matters” lines, and label what kind of change each one is. Code checks every quotation against the text we saved: the words must be there, in order, without a dropped “not”. That proves the words are in the source; it does not prove the person meant them as quoted. English sources only; X is not read yet.