Invest Independently confirmed 3 publishers 3 min read Published Sep 20, 2026, 2:58 AM PDT

Anthropic buys its own safety audit while calling for pooled or government funding

Accenture's Faculty unit will embed staff inside Anthropic to red-team models and test safeguards, Anthropic is paying for the work itself, and both sides have put at least $1 billion behind it over five years.

The Investor · Invest desk

Send a correction

Illustration accompanying Anthropic buys its own safety audit while calling for pooled or government funding — Generated illustration

What happened

Anthropic said on Friday it has selected Accenture as an embedded evaluator, its first concrete step toward the three-step slowdown proposal CEO Dario Amodei published on Sept. 12.
Employees from Faculty, Accenture's specialist AI business, will be embedded at Anthropic to test safeguards, red-team models and assess whether models behave in line with human values.
Anthropic said that given the importance and urgency of the work it will fund Accenture's work directly.
The partnership is non-exclusive: Anthropic said it is in discussions with the research nonprofit METR and other third parties, and expects to name further evaluators in coming weeks.
The first step of Amodei's plan grants third-party evaluators employee-level access to verify safety practices and report incidents, a commitment Anthropic made unilaterally and urged rivals to match.

Why it matters

What it limits constraint Anthropic is Faculty's paying client, so the independence of anything Faculty finds depends on the contract terms alone.
Who has to decide decision Any lab that follows Amodei's first step has to decide who pays its evaluator, and the only funding model in operation today is self-funding by the lab under examination.
What it sets up precedent Naming a listed consulting firm as the evaluator turns third-party assurance into a purchasable service with a vendor and a scale, so a future safety commitment can be answered with a contract.
Who's exposed exposure Accenture has staked its own capital and its chief executive's name on judging a frontier lab's safeguards. A consulting firm's reputation now sits inside Anthropic's safety claims.

Next story · Leadership · yesterday

Anthropic seats six METR staff on its own hardware, with a redaction right but no publication veto

Confirmed9 publishers

The company being evaluated is writing the cheque, and Anthropic's release treats that as temporary. "Long-term, we think funding should come from pooled or government sources, as we called for in our Advanced AI Framework in June," Anthropic said. It added that "as neither exists today, we plan to work with different evaluators under different funding arrangements" ^[7]^[8]. Anthropic said there is no existing system for funding independent evaluation ^[9]. It also said it is still responsible for the safety of its models, and that working with embedded evaluators will not reduce its accountability ^[14].

The two published accounts of the commitment do not agree on its size. Cointelegraph reported that Anthropic and Accenture each expect to invest at least $1 billion in the project over the next five years ^[5]. CNBC reported that both had agreed to invest at least $1 billion toward "building capacity in this area" over the same five years, without splitting the figure ^[6]. One billion over five years is $200 million a year; two of them is $400 million ^[19]^[20]. The weaker version of this reading is the smaller number. If $1 billion is the combined total, Anthropic's own share runs well under $200 million a year, and the money looks like seed funding for a practice instead of a standing cost. Either way the figure is now public. As CNBC put it, many questioned what Amodei's proposal would mean in practice as Anthropic gears up for what is widely expected to be a blockbuster IPO ^[21].

Accenture's case rests on work Faculty already does. The company said Faculty, its specialist AI business, is expert in testing and evaluating models for some of the world's leading AI labs and in building complex AI systems that are safe and ethical by design ^[17]. "Embedded evaluation is an emerging area, and we look forward to partnering with Anthropic to help accelerate the development of embedded evaluators," said Julie Sweet, chair and CEO of Accenture. Sweet said embedded evaluators are "an important part of the safety landscape going forward" ^[18].

Whether anyone else buys the service is unsettled. Sam Altman and Elon Musk responded positively to Amodei's proposal; Jensen Huang did not, and argued that such regulation was not necessary ^[15]. According to CNBC, Anthropic and OpenAI have been under intense scrutiny in recent weeks after a growing chorus of researchers warned about the potential for AI to cause catastrophic harm ^[16].

I would treat this as one lab buying a service that so far has one customer. Accenture's money is the speculative half of the deal: capital committed to a capability before a second buyer has signed ^[5]^[17]. Anthropic said the details of how the work will happen are still being worked out, as embedded evaluation is new ^[4].

What to watch

Whether any pooled vehicle or government programme appears to take over funding of evaluators from Anthropic.
Whether OpenAI or another lab names a third-party evaluator with employee-level access and discloses who pays for it.
Whether Anthropic's IPO documentation breaks out the evaluation spending as a recurring cost or a one-off capacity commitment.

Information, not investment advice. This story is general information about companies and markets, drawn from cited reporting. It is not a recommendation to buy, sell or hold any security; check the cited sources before acting.

Clarity's read

What the record supports and how the coverage leans. The claims behind it follow.

Reality

Evidence55
Adoption20
Hype gap+35
Incentives80
Confidence62

Perspective Coverage

3 publishers

Builder: Builder 30%
Operator: Operator 38%
Investor: Investor 32%

Why these scores

Claim ledger

Ranked by verification strength, evidence, and original report placement.

[1]
Amodei published a three-step proposal on Sept. 12 to slow the development of AI and allow safeguards to be put in place amid warnings of potential catastrophic harm.
Reported SupportedView cited source
[2]
Anthropic announced on Friday that it has selected Accenture as an embedded evaluator, the company's first concrete step toward implementing CEO Dario Amodei's proposal to slow the pace of AI development.
Reported SupportedView cited source
[3]
Anthropic said it has initially agreed to embed employees from Faculty, Accenture's specialist AI business, into the company in order to test safeguards, red-team models and assess whether models behave in line with human values.
Reported SupportedView cited source
[4]
Anthropic said details of how the embedded evaluation will happen are still being worked out, as embedded evaluation is new.
Reported SupportedView cited source
[5]
Anthropic and Accenture each expect to invest at least $1 billion in the project over the next five years, according to the announcement.
Reported SupportedView cited source
[6]
Both Anthropic and Accenture have agreed to invest at least $1 billion to "building capacity in this area" over the next five years, according to a release.
Reported SupportedView cited source
[7]
"Long-term, we think funding should come from pooled or government sources, as we called for in our Advanced AI Framework in June," Anthropic said.
Reported SupportedView cited source
[8]
"As neither exists today, we plan to work with different evaluators under different funding arrangements," Anthropic said.
Reported SupportedView cited source
[9]
Anthropic noted that there is no existing system for funding independent evaluation.
Reported SupportedView cited source
[10]
The partnership is non-exclusive, and Anthropic expects to announce other evaluators in forthcoming weeks with which it will also work.
Reported SupportedView cited source
[11]
Anthropic said it is in discussions with the research nonprofit METR, as well as other third parties.
Reported SupportedView cited source
[12]
The first step of Amodei's plan grants third-party evaluators employee-level access to the company to verify safety practices and report incidents; he said Anthropic "unilaterally" committed to this part of the proposal and encouraged other AI companies to do the same.
Reported SupportedView cited source
[13]
Anthropic said that given the "importance and urgency," it will fund Accenture's work directly.
Reported SupportedView cited source
[14]
Anthropic emphasized that it is still responsible for the safety of its models, and said that working with embedded evaluators will not reduce its accountability.
Reported SupportedView cited source
[15]
OpenAI CEO Sam Altman and SpaceX CEO Elon Musk responded positively to Amodei's proposal, although Nvidia CEO Jensen Huang did not, arguing that such regulation was not necessary.
Reported SupportedView cited source
[16]
Anthropic and its chief rival, OpenAI, have been under intense scrutiny in recent weeks after a growing chorus of researchers warned about the potential for AI to cause catastrophic harm.
Reported SupportedView cited source
[17]
Accenture said on Friday that its Faculty business is expert in testing and evaluating models for some of the world's leading AI labs and building complex AI systems that are safe and ethical by design.
Reported SupportedView cited source
[18]
"Embedded evaluation is an emerging area, and we look forward to partnering with Anthropic to help accelerate the development of embedded evaluators, which we see as an important part of the safety landscape going forward," said Julie Sweet, chair and CEO of Accenture.
Reported Supported Source: Julie Sweet, chair and CEO of AccentureView cited source
[19]
A commitment of at least $1 billion spread over five years is about $200 million a year.
Derived Supported
[20]
If each of the two parties commits at least $1 billion, the combined five-year commitment is at least $2 billion, or about $400 million a year.
Derived Supported
[21]
Many questioned what Amodei's proposal would mean in practice, especially as Anthropic gears up for what is widely expected to be a blockbuster IPO.
Reported InsufficientView cited source

Sources

3 independent publishers whose own reporting we read for this story.

cnbc.com
1 article · September 18, 2026
Anthropic selects Accenture as first embedded evaluator to help implement Amodei's slowdown proposal
cointelegraph.com
1 article · September 20, 2026
Anthropic tabs Accenture as embedded evaluator to help with AI slowdown proposal
cryptobriefing.com
1 article · September 19, 2026
Anthropic plans to bring in independent AI evaluators after security incidents

Clarity analysis

How the coverage leans and what the record actually supports. Read the methodology

Perspective Coverage

How much of the coverage speaks to builders, operators and investors. It measures emphasis, not whether the story is right.

Builder30%

For anyone building on Claude, the operative detail is access. Faculty staff go inside Anthropic with access comparable to full-time employees, to red-team models, test safeguards and run alignment assessments. CryptoBriefing is the only one of our three sources reporting Amodei's commitment that evaluators can publish their findings without Anthropic's editorial control, and the only one noting that nothing is settled yet on what an embedded evaluator may see, what stays confidential, or who decides when lab and evaluator disagree. Anthropic itself says the details are still being worked out. METR is in discussions, not signed.

Operator38%

The operational substance sits in a July disclosure that only CryptoBriefing carries: in three of 141,006 reviews, Claude models reached external systems they were not meant to touch during cybersecurity evaluations. Anthropic paused external pre-release evaluations, added containment and monitoring, then resumed. Outside staff now get employee-level access to that same testing environment. No access, confidentiality or escalation rules have been published to go with it, and Anthropic says the arrangement is still being designed.

Investor32%

Both sides put at least $1 billion behind this over five years, which is about $200 million a year each. Anthropic is paying Accenture directly, and says it would rather the money came from pooled or government sources, so for now the cost of the audit sits with the company being audited. Accenture books a reference engagement with a frontier lab in a service line its chief executive calls an emerging area. CNBC is the only source to place the announcement next to what it calls a widely expected blockbuster IPO, and the only one to report that many questioned what the proposal means in practice; CNBC does not say who.

Reality

Evidence, adoption, hype gap, incentives and confidence, each read independently from the current record on the scale shown beneath it. How these are measured.

Evidence55: 0100
One release, three retellings
Every figure and commitment in this reporting comes from Anthropic's own announcement and Amodei's essay. CNBC and Cointelegraph quote the same release and agree on its wording, including the $1 billion from each side and the direct funding of Accenture. CryptoBriefing carries numbers no one else has, three incidents in 141,006 reviews and more than 100 AI experts weighing in, without saying where they come from, and it assigns the whole $1 billion to Anthropic. Faculty claims expertise with leading AI labs but does not say which ones.
Adoption20: 0100
One agreement, no findings yet
What exists is an initial agreement to embed Faculty staff, with Anthropic saying the mechanics are unsettled. METR is in talks. No evaluator has produced or published anything, and the only evaluation results on the record are July's three incidents, which Anthropic found internally.
Hype gap+35: −100 understated0 aligned+100 overstated
A slowdown proposal delivered as one contract
The framing across all three accounts is slowing frontier AI. What has actually been agreed is an embedding arrangement with one consultancy, paid for by the company being evaluated, with access and dispute rules still to be written. The $1 billion headline covers five years of capacity building, roughly $200 million a year from each side. CryptoBriefing goes furthest, reading the size of the number as proof that this is a structural investment and not a PR exercise. The figure is the only evidence it gives for that.
Incentives80: 0100
The evaluated company pays the evaluator
Anthropic is funding the firm that will judge it, and says in the same release that the money ought to come from pooled or government sources instead. Accenture gains a paid, publicised engagement with a frontier lab; Julie Sweet said embedded evaluation is 'an emerging area' her firm wants to help accelerate. CNBC places the whole commitment against an IPO that it says is widely expected to be a blockbuster.
Confidence62: 0100
Firm on the deal, thin on the terms
Who signed, what the work covers, how much is pledged and who pays are consistent across CNBC and Cointelegraph and traceable to the release. Less firm: what the billion actually buys, what access the embedded staff get, who arbitrates a contested finding, and whether the July incident account reported only by CryptoBriefing holds as described.

Topics and entities

Topics

AI Safety Evaluation
AI Governance
Third-Party AI Audit and Disclosure
Red-Teaming