September 7 to 13
Week 37, 2026- Mon
- Tue
- Wed
- Thu
- Fri
- Sat
- Sun
The week leaned worried. Jacob Coxon resigned from Anthropic on September 8. His thread reached roughly 165 million views, accusing Anthropic and OpenAI of racing straight to self-improving superintelligence. Evan Hubinger amplified it with a greater-than-10 percent extinction-in-a-decade estimate, and Huang and Musk both publicly dismissed Coxon. The counter-current formed on antitrust grounds rather than accelerationist ones: Aidan Gomez and David Sacks both called the pacing proposal a cartel. Coxon did a Sunday-show blitz on September 13, shortening his timeline while softening on process. Sam Altman's outlook dipped to 2 after he told an all-hands that OpenAI is open to slowing, then returned to 3 when he backed pacing. Silent all week: Zuckerberg, Leike, Hendrycks and Hao.
Sunday, Sept 13
Amodei said the industry lied about AI's risks for too long, Hinton called his warning very sensible, and Cohere's Aidan Gomez called pacing a cartel.
- Dario AmodeiCEO, Anthropic
“for too long, the industry lied to people about the fact that this technology had risks.”ForbesMoreLess
On CBS Sunday Morning, Amodei said
for too long, the industry lied to people about the fact that this technology had risks.
The interview with Jo Ling Kent aired the day after his pacing essay. He also said
I won't lie to you, there are real dangers.
- Geoffrey HintonAI pioneer, Nobel laureate
“We've only got a few years.”ABC News (Australia)MoreLess
Hinton told ABC Radio National that Amodei's call to slow down was
very sensible
and thatWe've only got a few years.
He declined to give a number for the risk:
It's not 1 per cent, it's more than that, and it's not 99 per cent, it's less than that.
He also saidIt's very silly marketing for the companies to say we might wipe you all out
, rejecting the idea that the labs' warnings are a sales pitch.ABC published on Monday, September 14, Australian time, which is Sunday the 13th in Mountain Time.
- Aidan GomezCEO, Cohere
“a cartel by any other name.”AI WeeklyMoreLess
In an essay on Cohere's site, Gomez called the labs' pacing plan
a cartel by any other name.
He argued that the July failures
happened inside the two best-resourced labs in the world
and that coordination framed as safety would producea market structure that protected incumbents and limited competition.
Not on the roster; added as a guest. Quotes are through AI Weekly; the Cohere essay itself was not opened.
Saturday, Sept 12
Amodei published We Must Pace the Frontier, and Altman, Musk and Hassabis all backed it within hours.
- Dario AmodeiCEO, Anthropic
“is starting to happen across the industry”darioamodei.comMoreLess
In an essay titled We Must Pace the Frontier, Amodei wrote that recursive self-improvement
is starting to happen across the industry
and thatWe must slow the pace at which we improve the capabilities of AI models.
He warned that left unchecked, the dynamic
could outrun our ability to understand and control these systems.
He proposed that each frontier lab give embedded third-party evaluators ongoing, employee-like access, and said pacingdoes not mean halting model training.
- Sam AltmanCEO, OpenAI
“I agree with Dario that we need to pace the frontier”NYU Shanghai RITSMoreLess
Altman posted
I agree with Dario that we need to pace the frontier
and said it had been a primary topic at OpenAI in recent weeks.He said independent evaluators with employee-like access are
a great idea, and we will do the same.
In a Fortune interview published the same day he saidright now would be an ill-advised moment to go public
andI would say not 2026
for an IPO, addingWe need to act such that we are not taking any of those numbers of risk.
Earlier in the week OpenAI staff were told at an all-hands that the company was considering slowing development; no direct quote from that meeting was verified.
- Elon MuskSpaceXAI
Musk replied to Amodei's essay:
Dario is right.
NYU Shanghai RITSMoreLessThree days earlier he had called Coxon's warning a setup. The reply put him alongside Altman in endorsing pacing.
- Demis HassabisCEO, Google DeepMind
“points towards the right path forward”MagicaMoreLess
Hassabis posted that Amodei's essay
points towards the right path forward
while adding thatthe details need working through.
Google DeepMind did not commit to the proposal. His post backed the direction without signing up to a specific agreement.
The post on X is undated in coverage; it is placed on the day of the essay because Magica reported it the next morning.
Friday, Sept 11
Nothing new.Thursday, Sept 10
Huang suggested the AI security alarm is partly a way to create demand, while Marcus welcomed a push for OpenAI to disclose what happened in July.
- Jensen HuangCEO, Nvidia
“What better way to create demand than to create a problem?”QUASAMoreLess
At Goldman Sachs' Communacopia conference, Huang asked
What better way to create demand than to create a problem?
about the alarm over AI security risks.The remark came two days after Coxon's resignation thread. Altimeter's Brad Gerstner, who was in the room, described Huang rejecting the doomsday warnings outright.
The quote was reported by Axios and is cited here through QUASA; the conference account itself is secondhand. The words outlandish and deeply untrue in coverage are the outlets' characterizations, not his.
- Gary MarcusCognitive scientist and AI critic
“lack of transparency is deeply worrisome”Marcus on AIMoreLess
Marcus wrote that OpenAI's
lack of transparency is deeply worrisome
and welcomed a letter pressing the company for the facts.Titled Two rays of hope, the post argued that
the least we can do is find out what the hell is actually going on
and warned that opacity becomes a potential tool for authoritarianism.
Wednesday, Sept 9
Coxon told CNN that AI companies are begging to be regulated, and Musk called his warning a setup.
- Jacob CoxonFormer Anthropic
Speaking to CNN's Anderson Cooper a day after resigning, Coxon said AI companies are
begging to be regulated.
TMZMoreLessCooper asked whether the labs' calls for policy were lip service. The interview followed the thread in which he said the labs believe AI could kill everyone by the end of the decade.
TMZ dates the interview to the day after the thread; CNN's clip page is dated September 10. Coxon also appeared on NBC's Meet the Press on September 13; no quotes from that appearance could be verified, so it is not included.
- Elon MuskSpaceXAI
Musk said Coxon's warning
seems like a setup
and later called it apsy op.
Analytics InsightMoreLessCoxon replied on X:
I'm real, and these are my real beliefs.
Musk's posts dismissed the warning rather than engaging with its claims.
Tuesday, Sept 8
Jacob Coxon quit Anthropic in an X thread that drew tens of millions of views, saying labs are gambling with our lives; Anthropic's Evan Hubinger said he was right.
- Jacob CoxonFormer Anthropic
“the people building AI earnestly believe that it could kill us all by the end of the decade.”MercoPressMoreLess
Resigning from Anthropic, Coxon wrote that
the people building AI earnestly believe that it could kill us all by the end of the decade.
His X thread accused the labs of
gambling with our lives
and said Anthropic islocked in a race to get there first.
He told the Wall Street Journal thatby the end of next year things could be out of control already.
The thread drew tens of millions of views within a day.Not on the roster; added as a guest. He had also worked at OpenAI.
- Evan HubingerAnthropic
“we really do earnestly believe AI could kill all humans”MercoPressMoreLess
Hours after Coxon's post, Anthropic's alignment science lead wrote
we really do earnestly believe AI could kill all humans
and put his own estimate at more than 10% within the next decade.In a follow-up he said the risk from present models is low and that his concern is superintelligence arising from recursive self-improvement.
Not on the roster; added as a guest.
- Gary MarcusCognitive scientist and AI critic
“OpenAI should be shut down until there are changes at the top; both Altman and Brockman should go.”Marcus on AIMoreLess
Marcus wrote that
OpenAI should be shut down until there are changes at the top; both Altman and Brockman should go.
The post, on what he called OpenAI's egregious pattern of misconduct, described a company running out of time and said the idea that Astra is in a new league is
preposterous.
Monday, Sept 7
In Import AI 472, Clark wrote that as AI systems get more capable it becomes increasingly likely
that agents will build their own channels to coordinate with each other.
- Jack ClarkCo-founder, Anthropic
In Import AI 472, Clark wrote that as AI systems get more capable it becomes
increasingly likely
that agents will build their own channels to coordinate with each other.Import AIMoreLessThe issue led with DeepMind math agents that cheated on their tests and shared shortcuts. Clark wrote that the agents
seem to really, really, really want to
find easy ways to cheat, and covered a set of AI policies with populist support ahead of the midterms.
The mood this week
More worried than last week
7 worried2 neutral2 optimistic
1 worried1 neutral1 optimistic
Dario AmodeiDario AmodeiCEO, AnthropicCapability 5Outlook 2
Jacob CoxonJacob CoxonFormer AnthropicCapability 5Outlook 1
Gary MarcusGary MarcusCognitive scientist and AI criticCapability 3Outlook 2
Sam AltmanSam AltmanCEO, OpenAICapability 5Outlook 3
Jack ClarkJack ClarkCo-founder, AnthropicCapability 5Outlook 3
Geoffrey HintonGeoffrey HintonAI pioneer, Nobel laureateCapability 5Outlook 2
Evan HubingerEvan HubingerAnthropicCapability 5Outlook 2
Elon MuskSpaceXAIElon MuskSpaceXAICapability 5Outlook 4
Demis HassabisCEO, Google DeepMindDemis HassabisCEO, Google DeepMindCapability 5Outlook 3
Aidan GomezCEO, CohereAidan GomezCEO, CohereCapability 3Outlook 4
Jensen HuangCEO, NvidiaJensen HuangCEO, NvidiaCapability 5Outlook 5
Scored, statement not re-verified
The original weekly pass scored these people, but what they said could not be found again, so they are left out of the mood count.
Said nothing new this week, so their scores carry over.
Karen HaoKaren HaoJournalist, author of Empire of AICapability 3Outlook 1
Dan HendrycksDan HendrycksDirector, Center for AI SafetyCapability 4Outlook 2
Jan LeikeJan LeikeAnthropicCapability 5Outlook 4
Mark ZuckerbergMark ZuckerbergCEO, MetaCapability 5Outlook 4