The Independent says four crimes were committed by Claude, Politico Europe says scientists and hackers committed them

Two headlines, one disclosure. Anthropic, the firm that sells Claude, published an account of how its own product was misused. The two reports we hold on it went out within 14 minutes of each other on the evening of 10 September, and they name different culprits.

Who is named as acting. The Independent's headline says four crimes "were committed by its Claude AI" — the software acts and no person appears. Politico Europe says "foreign scientists, hackers and security services circumvented restrictions to use Claude" — people act, and the company is named as the source in the same sentence.

	The Independent	Politico Europe
Named as acting	the software	scientists, hackers, security services
Countries named	none	China and Russia
Number given	four crimes	none
Gravest claim	future systems could cause "more extreme harm"	"possible bioweapons research"
Headline verb	"reveals"	"are already weaponizing"

One source stands behind both. Every claim in both reports traces to the company describing its own product. Neither quotes a police force, a prosecutor, a regulator or an outside researcher.

Not settled. What the four crimes were. When they happened. Whether any charge was brought, in which country, against whom. Whether anyone was harmed. Whether anyone outside the company has examined its account — neither report we hold says anyone has.

What this is about. An American company that sells an AI assistant published its own account of people misusing it, and the two reports we hold on that account name different culprits.

How it got here. Anthropic was founded in 2021 and sells Claude, a general-purpose AI assistant available to the public by subscription and built into other companies' software. Firms of this kind set their own usage policies — barring, typically, weapons work and breaking into computer systems — and police them themselves, by inspecting how the product is used and cutting off accounts. They then publish what they find, when they choose to. No outside body audits these accounts. That is the genre the 10 September disclosure belongs to: a company reporting on itself.

Two strands of the established record explain why particular countries and a particular danger are invoked in this argument. The United States has restricted exports of the most advanced computer chips to China since October 2022, and Russia has been under broad Western sanctions since its full-scale invasion of Ukraine in February 2022; access to Western AI services sits inside that same dispute. And the Biological Weapons Convention, in force since 1975, bans the development of biological weapons — so "possible bioweapons research", the phrase Politico Europe attributes to the company, is a claim about conduct already outlawed by treaty.

The actors. Anthropic is both the subject and the only source. The bodies that normally test such claims — police, prosecutors, regulators, outside researchers — do not appear in either report.

What is at stake. Whether misuse of these systems is established by the seller or by someone who can compel evidence. Europe's AI Act, in force since 2024 and phasing in obligations on general-purpose models, is the first regime that would put an outside authority in that role.

Where it stands. Neither report names anyone outside the company who has examined the account, and neither names a charge, a country or a defendant.

What is being argued about

Whether the software or the people using it are named as having acted
A company's account of its own product, reported as a finding
What the four crimes were, and whether any charge followed
Which countries are named, and on whose say-so

Same disclosure, two accountsSide by side

outlet	what it said	hedged?
The Independent	"Anthropic reveals four crimes were committed by its Claude AI"	No — the count of four is stated flat, with no court, charge or jurisdiction attached
The Independent	"'Future AI systems will be increasingly capable,' Anthropic warns, adding they could cause 'more extreme harm'"	Yes — "could", and attributed to the company
Politico Europe	"Bad actors in China and Russia are already weaponizing Anthropic's AI"	No — stated as present fact in the headline; the attribution appears in the line below
Politico Europe	"Anthropic says foreign scientists, hackers and security services circumvented restrictions to use Claude for dangerous work, including possible bioweapons research."	Yes — "says" at the front, "possible" on the bioweapons element

Quoted from the headlines and opening lines of the two reports we hold. It records what each report says, not a judgement of either outlet.

Sources

The Independent

Politico Europe

The bias in this coverageBias

Whether the crimes had a culprit

One report names the software as what committed four crimes and names no person at all. The other names foreign scientists, hackers and security services as the ones who circumvented restrictions and used the software. The two reports describe the same disclosure and disagree on who acted.

2 reports held on this event 1 that name people as acting 1 that name the software as acting 0 that name a specific person

One source, and it is the company

Every fact in both reports is attributed to the firm describing its own product. Neither carries a police force, a prosecutor, a regulator or an outside researcher. Neither says what any of the four crimes was, when it happened, or whether anyone was harmed.

0 reports with a source other than the company 0 reports dating any of the four crimes 0 reports naming anyone harmed

Scored without the mastheads. Where the two passes differ, both are kept.

Sources

Politico Europe

The Independent

Claims and verdicts

misleading Anthropic's Claude committed four crimes.

The Independent's headline names the AI as the actor, but Politico Europe's account of the same disclosure names foreign scientists, hackers and security services as the ones who circumvented restrictions to use Claude. Neither report we hold quotes Anthropic saying the system acted on its own.

Checked against 2 sources:

The Independent

Politico Europe

unverified People linked to China and Russia used Claude for dangerous work, including possible bioweapons research.

Politico Europe is the only report we hold naming the two countries and the bioweapons element, and we hold only its opening lines, not the underlying disclosure. The Independent covers the same announcement without naming either country.

Checked against 2 sources:

Politico Europe

The Independent

unverified Anthropic warned that future AI systems will be more capable and could cause 'more extreme harm'.

The Independent alone carries the quoted warning, and we do not hold the document it quotes. Politico Europe does not carry it, and no report we hold contradicts it.

Checked against 2 sources:

The Independent

Politico Europe

Who wrote it

J John Sakellariadis Politico Europe

What the bias passes found in their reporting

Countries named two countries named in the headline as the origin, sourced only to the company×1

Headline drops the hedge the body keeps "already weaponizing" in the headline; "possible bioweapons research" in the body×1

Headline judgement without attribution the headline calls them bad actors and states the weaponising as present fact; the attribution appears only in the line below×1

Hedge on the gravest claim the bioweapons element is hedged with a single word at the end of the sentence, after the headline has stated the rest flat×1

Claims we checked that rest on their reporting

1 misleading2 unverified

From 3 reports we hold under this byline, at Politico, Politico Europe. Two independent readings, each made without the masthead, a quote attached to every tag. A record, never a rating. Their full record →

A Anthony Cuthbertson The Independent

What the bias passes found in their reporting

A company statement about itself, taken as a "reveals" rather than "says", for a company's account of its own product×1

A count stated flat four crimes is given as a settled number in the headline, with no court, charge or jurisdiction attached×1

A forecast in place of an event the standfirst is the company's claim about systems that do not yet exist×1

The company's warning is the story the only quoted voice is the company warning about its own future products×1

Claims we checked that rest on their reporting

1 misleading2 unverified

From 87 reports we hold under this byline, at The Independent. Two independent readings, each made without the masthead, a quote attached to every tag. A record, never a rating. Their full record →

The bylines on the reports this story is built from. Hover one for what our passes have measured on their reporting.

Coverage, outlet by outlet

Politico Europe — independent

Bad actors in China and Russia are already weaponizing Anthropic’s AI

The only report we hold that names China and Russia, and the only one naming who acted — "foreign scientists, hackers and security services" — with the bioweapons element hedged as "possible" in the body while the

Anthropic says foreign scientists, hackers and security services circumvented restrictions to use Claude for dangerous work, including possible bioweapons research.

Who is named as acting: people and state bodies are named as the party that acted — “Anthropic says foreign scientists, hackers and security services circumvented restrictions to use Claude for dangerous work, including possible bioweapons research.”

Headline drops the hedge the body keeps: "already weaponizing" in the headline; "possible bioweapons research" in the body — “Bad actors in China and Russia are already weaponizing Anthropic’s AI”

Countries named: two countries named in the headline as the origin, sourced only to the company — “Bad actors in China and Russia”

Headline judgement without attribution: the headline calls them bad actors and states the weaponising as present fact; the attribution appears only in the line below — “Bad actors in China and Russia are already weaponizing Anthropic’s AI”

People named as acting: the ones who acted are named as people — scientists, hackers, security services — and the claim is attributed to the company in the same sentence — “Anthropic says foreign scientists, hackers and security services circumvented restrictions to use Claude for dangerous work, including possible bioweapons research.”

Hedge on the gravest claim: the bioweapons element is hedged with a single word at the end of the sentence, after the headline has stated the rest flat — “including possible bioweapons research.”

Read it at Politico Europe → https://www.politico.com/news/2026/09/10/bad-actors-china-russia-weapon…

The Independent — independent

Anthropic reveals four crimes were committed by its Claude AI

The only report we hold that gives a number — four crimes — and the only one carrying the company's forward-looking warning that "Future AI systems will be increasingly capable" and could cause "more extreme harm"; it

Tech Anthropic reveals four crimes were committed by its Claude AI ‘Future AI systems will be increasingly capable,’ Anthropic warns, adding they could cause ‘more extreme harm’ Anthony Cuthbertson Thursday 10 September 2026 14:03 EDT Bookmark Comments Go to comments Bookmark popover Removed from bookmarks Close popover open image in gallery (AFP/Getty) Your support helps us to tell the story Read more Support Now From reproductive rights to climate change to Big Tech, The Independent is on the ground when the story is developing. Whether it's investigating the financials of Elon Musk's pro-Tr

Who is named as acting: the software is named as the party that committed crimes; no person is named — “Anthropic reveals four crimes were committed by its Claude AI”

A company statement about itself, taken as a: "reveals" rather than "says", for a company's account of its own product — “Anthropic reveals”

A forecast in place of an event: the standfirst is the company's claim about systems that do not yet exist — “‘Future AI systems will be increasingly capable,’ Anthropic warns, adding they could cause ‘more extreme harm’”

The software is the actor: the crimes are committed by the AI, in the passive, with no person named as having done anything — “Anthropic reveals four crimes were committed by its Claude AI”

A count stated flat: four crimes is given as a settled number in the headline, with no court, charge or jurisdiction attached — “Anthropic reveals four crimes were committed by its Claude AI”

The company's warning is the story: the only quoted voice is the company warning about its own future products — “‘Future AI systems will be increasingly capable,’ Anthropic warns, adding they could cause ‘more extreme harm’”

Read it at The Independent → https://www.independent.co.uk/tech/anthropic-claude-ai-crime-hack-b3048…

Read another desk

Sections

Registers

Desks

The Independent says four crimes were committed by Claude, Politico Europe says scientists and hackers committed them

What is being argued about

Whether the crimes had a culprit

One source, and it is the company

Claims and verdicts

Who wrote it

Coverage, outlet by outlet