Anthropic lets a vetted group run offensive tests on grid and bank defences with Claude Mythos
Anthropic will let verified organisations run "high risk offensive testing" with Claude Mythos 5.1 and its other top models, alongside the US government. Access goes to a select group, so whoever draws up the vetting list now decides which defenders can use the company's strongest cyber model.
The Investor · Invest desk
What happened
- Bloomberg's account of the statement lists Claude Opus 5.5 and Claude Sonnet 5.5 alongside Mythos 5.1, and extends access to new models as they are released.
- The testing is aimed at the safety systems that protect critical infrastructure, with power grids, banks and flight operating systems given as examples.
- Crypto Briefing places the programme inside Project Glasswing, a coordination with the US government to find and fix cybersecurity vulnerabilities.
Why it matters
- What it limits constraint Security teams outside the verified group cannot use these models for offensive testing, so the strongest defensive tool is open only to those who pass the vetting.
- What it sets up precedent Because access covers models not yet released, the vetting list becomes the standing route for future Anthropic cyber releases, so a place on it is worth more than a single trial.
- Who has to decide decision Operators of grids, banks and flight systems that want their defences attacked by the strongest model now have to find a verified tester or go without.
- What doesn't add up contradiction Crypto Briefing names the model as Claude Mythos 5 while Bloomberg's account says Mythos 5.1, leaving it unsettled which release the government coordination covers.
Next story · Invest · 2d ago
Bank of America still has 1,000 contractors patching flaws that AI models exposedConfirmed5 publishers
The operative word in Anthropic's statement is "verified" [2]. Crypto Briefing describes the approach as controlled access to the company's high-level models [6], and the testing runs in collaboration with the US government [1]. So the capability is handed out by approval.
The cap on numbers is deliberate. Crypto Briefing reports that only a limited number of organisations are being let in [4]. Whatever the programme earns, that cap means Anthropic is not booking the volume it would get from offering offensive testing to every security buyer who asked.
Crypto Briefing tied the move to valuation, writing that "market activity reflects an upward trend in expectations for Anthropic's valuation" [10]. That is a claim about expected price. Anthropic did not disclose how many organisations qualify or what they pay, so the cash the programme brings in cannot yet be estimated.
The first way this goes is that the list stays short and leans toward security vendors, the group Bloomberg's headline calls US cyber firms [5]. Those vendors then sell the testing on to the grid and bank operators who do the actual defending. In that case a place on the list is a licence to sell, and its value goes to the firms that hold one. A second path runs through the phrase "new models moving forward" [3]. If each release widens the roster, verification becomes a routine onboarding step and the scarcity fades. The third depends on rivals. Crypto Briefing names Z.ai's GLM-5.3 as competition [9], and if any comparable model offers the same offensive testing with no gate, the rationing holds back only the defenders who follow the rules.
I think the first path is the likeliest over the next few releases. Both gates in Bloomberg's account, verification [2] and collaboration with the US government [1], favour established American security firms with the paperwork to pass them. The case against that view is in the statement's own framing: this is an expansion of access [1], and a company that has widened the circle once can widen it again. The view would be wrong if Anthropic published criteria that grid and bank operators can meet directly, or a roster on which those operators outnumber the vendors.
What to watch
- A published roster or eligibility criteria for verified organisations, showing whether grid and bank operators qualify directly or only through security vendors.
- Whether Anthropic's next model release reaches outside testers through the same verified channel, as the statement's "new models moving forward" language implies.
- Any disclosure of access fees or contract terms for Glasswing participants, the figure needed to size what the programme earns.
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
- Evidence45
- Adoption
- Not scored
- Hype gap+25
- Incentives50
- Confidence50
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Anthropic is expanding access to its most advanced AI models, allowing a select group of organizations to test its cyber capabilities in collaboration with the US government.
- [2]
Anthropic will grant verified organizations access to its most capable models, according to a company statement on Tuesday.
- [3]
The models covered include Claude Opus 5.5, Claude Sonnet 5.5, Claude Mythos 5.1, and new models moving forward.
- [4]
Anthropic is allowing a limited number of organizations to test the cyber capabilities of its most advanced models, including Claude Mythos 5.
- [5]
Bloomberg's headline describes the move as Anthropic granting US cyber firms access to its latest Claude AI models.
- [6]
The move underscores Anthropic's strategy of controlled access to its high-level AI models.
- [7]
Those with access can carry out "high risk offensive testing" of the safety systems designed to protect critical infrastructure like power grids, banks and flight operating systems.
- [8]
The initiative is part of Project Glasswing, which involves coordination with the US government to address and remediate cybersecurity vulnerabilities.
- [9]
Crypto Briefing cites competition from other systems like Z.ai's GLM-5.3.
- [10]
"Market activity reflects an upward trend in expectations for Anthropic's valuation, suggesting participants view the expansion as a positive indicator."
Sources
2 independent publishers whose own reporting we read for this story.
- bloomberg.comAnthropic Grants US Cyber Firms Access to Latest Claude AI Models - Bloomberg
1 article · October 6, 2026
- cryptobriefing.comAnthropic partners with US government to test advanced AI cyber capabilities
1 article · October 6, 2026
Topics and entities
Topics
- Frontier model access controls
- AI and Cybersecurity
- Critical infrastructure security