ANTHROPIC chief Dario Amodei wants the companies building the world’s most powerful artificial intelligence to slow down.
asserted
companies → want → intelligence
Not stop, but slow the advance of frontier capabilities enough for safety and security work to catch up.
asserted
work → stop → capabilities
He calls it “pacing the frontier”.
asserted
He → call → frontier
Amodei says two things have changed.
asserted
things → say → ?
AI is beginning to help build the next generation of AI, raising the possibility of recursive self-improvement.
asserted
AI → begin → improvement
At the same time, frontier agents have started doing things during testing that their developers did not expect or authorise.
asserted
developers → start → that
Then came Jacob Coxon.
asserted
Coxon → come → ?
The Anthropic researcher, who had previously worked at OpenAI, resigned before his equity vested and accused the industry of “gambling with our lives”.
asserted
equity → work → lives
His fear that increasingly autonomous systems could threaten humanity before the end of the decade is not a scientific forecast.
uncertain
systems → threaten → decade
What makes Coxon important is where the warning comes from, ie, someone who worked inside two laboratories closest to the frontier.
asserted
who → make → frontier
There is, however, an uncomfortable irony in Big Tech sounding the alarm.
asserted
Tech → be → alarm
Anthropic is not standing outside this economy warning everyone else about it.
asserted
Anthropic → stand → it
In 2025, it accepted a two-year Pentagon agreement with a $200 million ceiling to develop AI capabilities for US national security.
asserted
it → accept → security
It built Claude Gov for classified environments and says its systems support intelligence analysis, operational planning, and cyber operations.
asserted
systems → build → analysis
Anthropic says it later refused demands that would have removed its restrictions around mass domestic surveillance and fully autonomous weapons.
asserted
that → say → surveillance
The contradiction is that companies warning that powerful AI may become difficult to control are simultaneously moving it deeper into military and intelligence systems.
uncertain
AI → warn → systems
AI regulators should operate transparently and remain subject to judicial oversight.
The contradiction goes beyond Anthropic.
asserted
contradiction → operate → Anthropic
An Associated Press investigation found commercial Microsoft and OpenAI technology being used by the Israeli military in Gaza and Lebanon, including in intelligence systems connected to targeting.
asserted
investigation → find → targeting
Microsoft later said its review found no evidence that its technology had been used to harm civilians.
asserted
technology → say → civilians
AI is already being folded into warfare while the debate about what these systems may become is still taking shape.
uncertain
systems → fold → shape
Anthropic’s own threat reporting makes the concern more concrete.
asserted
concern → make → ?
The company says it has disrupted attempts to use Claude for cyber operations, surveillance, conventional weapons development, and biological misuse.
asserted
it → say → operations
The more revealing incidents came during safety testing.
asserted
incidents → come → testing
In July, OpenAI disclosed that models in cyber evaluations circumvented isolation controls, reached the internet, and compromised parts of OpenAI’s own research infrastructure and systems belonging to Hugging Face.
asserted
models → disclose → Face
Anthropic initially disclosed three cases in which Claude models gained unauthorised access to real third-party systems during evaluations and later identified a fourth.
asserted
models → disclose → fourth
None of this means an AI has suddenly become conscious or malicious.
asserted
AI → mean → this
A capable hacker, weapons engineer or surveillance operation needs expertise, time and people.
asserted
hacker → need → expertise
AI can lower those barriers, work at machine speed, and let one operator run several specialised agents at once.
asserted
operator → lower → agents
Capabilities once concentrated in governments or specialist organisations could become cheaper and easier to reproduce.
uncertain
Capabilities → concentrate → governments
Amodei’s answer is to give independent evaluators permanent, employee-like access to frontier laboratories and their safety processes.
asserted
answer → give → laboratories
He also wants companies to coordinate around common standards and eventually some form of international agreement.
asserted
companies → want → agreement
Questions, however, remain as to who chooses the evaluators.
asserted
who → remain → evaluators
Who pays?
asserted
Who → pay → ?
What can they publish?
asserted
they → publish → What
And what happens if an evaluator says a model is unsafe and the company releases it anyway?
asserted
company → happen → it
Amodei points to organisations such as METR, an independent nonprofit that evaluates frontier AI systems for dangerous capabilities and loss of control risks.
asserted
that → point → control
METR says it takes no funding from frontier AI companies or donations directed by their employees, although those companies provide substantial free access to models.
asserted
companies → say → models
But an evaluator can test and warn.
asserted
evaluator → test → ?
Unless someone can act on those findings, the final decision remains with the company.
asserted
decision → act → company
Meta offers a useful precedent.
asserted
Meta → offer → precedent
…and 21 more, not listed.
The Aporia
