MythosWatch

Updates

Latest source-backed changes.

Newest first. Each entry captures what changed, who it affects, source links, and why it matters for the distribution of AI capability.

Critical
AnthropicOfficial

Bloomberg: a small group of unauthorized users accessed Claude Mythos Preview

Bloomberg News, citing documentation and a person familiar with the matter, reported that a small group of unauthorized users gained access to Anthropic's restricted Claude Mythos Preview model.

An unauthorized access incident directly challenges Anthropic's Project Glasswing gating strategy and raises questions about the robustness of access controls on a model its own safety team flagged as too dangerous for public release.

View entity profile
Bloomberg / MarketScreenerwire
High
Australia ASIC and APRARegulatory watch

Asia-Pacific financial regulators monitor Mythos-related cyber risk

Reuters reporting carried by iTnews named ASIC, APRA, HKMA, South Korean financial regulators, and Singapore MAS as monitoring or responding to Mythos-related financial-system risk.

The access story is now also a reaction story: regulators without public access are still adjusting supervision and resilience planning around the model's reported capability.

View entity profile
iTnews / Reutersreported
High
Cybersecurity and Infrastructure Security AgencyBlocked

Axios: CISA lacks access to Mythos despite being the nation's top cyber defense agency

Axios reported that CISA does not have access to Claude Mythos Preview, even as other government agencies including the NSA are reportedly using it. Anthropic briefed CISA on Mythos capabilities but this did not result in access; the Commerce Department's Center for AI Standards and Innovation was separately reported to be testing the model.

The country's top cyber defense agency being excluded from a model it is supposed to help protect against creates a direct accountability and preparedness gap in U.S. critical infrastructure defense.

View entity profile
AxiosreportedAxiosreported
High
Pentagon / Department of DefenseIn discussion

CNBC: Trump says Anthropic is 'shaping up' and a DoD deal is 'possible'

CNBC reported that President Trump told reporters a deal between Anthropic and the Department of Defense was 'possible,' citing 'very good talks' following the White House meeting with Anthropic CEO Dario Amodei. The Pentagon had blacklisted Anthropic as a supply-chain risk in March 2026 after contract disputes over autonomous weapons and domestic surveillance.

A direct presidential signal that a DoD deal is possible changes the Pentagon's status from administratively blocked to active discussion at the executive level, with significant implications for military AI procurement.

View entity profile
CNBCreported
High
Infrastructure and governance exposure indicatorsExposure indicator

MythosWatch adds infrastructure and governance exposure indicators

The dataset separates contextual capacity indicators from direct access, using cyber-capacity, infrastructure, and governance sources.

Capability distribution is not only a list of users. Infrastructure concentration, cyber readiness, and governance participation shape who benefits and who must adapt.

View entity profile
World Economic ForumofficialAfrica Data Centres AssociationanalysisTechAfrica NewssecondaryUNECA / WSIS ForumanalysisAssociated PresswireDigital Watch Observatoryanalysis
Medium
Singapore MAS and Cyber Security AgencyRegulatory response

Singapore and Hong Kong add financial-regulator response signals

MAS warned financial institutions to strengthen defenses while coordinating with Singapore's cyber agency; HKMA reportedly engaged banks and planned a public-private AI cyber-risk taskforce.

Small financial hubs can be strategically important even when they are visually small on a world map.

View entity profile
iTnews / Reutersreported
High
National Security AgencyReported

TechCrunch amplifies Axios report that NSA is using Mythos despite the Pentagon feud

A secondary report summarized Axios reporting that the NSA is allegedly using Anthropic's restricted model while the Department of Defense remains in conflict with Anthropic.

This is the cleanest viral contradiction: one part of the security state reportedly uses the model while another part challenges the vendor.

View entity profile
TechCrunchsecondaryAxiosreported
Critical
National Security AgencyReported

Axios reports NSA use of Anthropic's Mythos

Axios reported that the National Security Agency is using Claude Mythos Preview despite a Defense Department effort to treat Anthropic as a supply-chain risk.

Reported intelligence use changes Mythos from a corporate cybersecurity preview into a state-capability access story.

View entity profile
Axiosreported
High
Pentagon / Department of DefenseBlocked

AP details the Anthropic and Pentagon dispute over allowed government uses

AP reported that the administration and Defense Department had sought to restrict federal use of Anthropic products amid conflict over autonomous weapons and domestic surveillance constraints.

Denied or contested access is just as important as granted access because it shows where policy, safety, and procurement collide.

View entity profile
Associated Presswire
High
OMB / Civilian Federal AgenciesPrepared

OMB reportedly prepares guardrails for possible federal-agency Mythos access

Reporting based on a Bloomberg-reviewed memo says OMB was setting up protections that could let agencies begin using a modified version of Mythos, without committing agencies or timing.

This is the clearest signal of a civilian-government access lane forming alongside the Pentagon dispute.

View entity profile
ForbessecondaryCSO Onlinesecondary
High
White HouseIn discussion

White House chief of staff meets Anthropic CEO about Mythos

AP reported that White House chief of staff Susie Wiles met with Anthropic CEO Dario Amodei to discuss Mythos, cybersecurity, AI safety, and U.S. leadership.

The meeting makes Mythos access an executive-branch policy issue and creates a credible basis for tracking government discussions separately from confirmed deployments.

View entity profile
Associated Presswire
High
European Central Bank and eurozone lendersRegulatory watch

ECB reportedly convenes eurozone bank risk officers on Mythos

Bloomberg reported that the European Central Bank was convening chief risk officers of eurozone lenders to discuss potential threats from Anthropic's Mythos model.

Europe's financial regulators appear to be moving even where direct testing access remains unclear.

View entity profile
Bloombergreported
Medium
White HouseIn discussion

Axios reports active U.S. government discussions over Mythos access

Axios reported that White House officials were weighing access to Anthropic's new Mythos technology despite the Pentagon conflict.

This establishes a pre-meeting signal that federal access discussions were already moving before public confirmation of the White House meeting.

View entity profile
Axiosreported
High
South Korea MSIT and financial regulatorsRegulatory response

South Korea holds emergency cybersecurity briefings after Mythos shock

Seoul Economic Daily reported that South Korea's science ministry met with major domestic cybersecurity firms to discuss countermeasures after specialized cyber AI releases.

This is a named ministry response, not a generalized sentiment signal, making it one of the strongest non-access records in Asia.

View entity profile
Seoul Economic Dailyreported
High
European Cyber AgenciesNo public access

European oversight bodies remain outside the named Mythos testing list

The Decoder, citing POLITICO reporting, says most European cyber agencies had limited visibility while Germany's BSI had talks without direct testing access and the UK AISI had evaluated the model.

The access pattern is not only national; it runs through institutions with different technical credibility and vendor relationships.

View entity profile
The Decoderanalysis
High
UK AI Security InstituteEvaluated

UK AISI publishes official evaluation of Claude Mythos Preview's cyber capabilities

The UK AI Security Institute published its official evaluation of Claude Mythos Preview's performance on cybersecurity tasks, including capture-the-flag challenges and complex multi-step attack simulations, confirming direct government evaluation of the model.

This primary source confirms the UK government is among the most technically informed allied governments on Mythos capabilities, and places the UK evaluation on record with official documentation rather than only secondary reporting.

View entity profile
UK AI Security Instituteofficial
Medium
Bank of England CMORGRegulatory watch

Bank of England CMORG reportedly prepares Mythos discussion with banks

Bloomberg reporting carried by Insurance Journal said Mythos would be on the agenda for the Bank of England's cross-market operational resilience and AI taskforce discussions.

Financial resilience meetings give the UK a second public lane beyond model evaluation: sector-wide preparedness.

View entity profile
Insurance Journal / Bloombergreported
Medium
Bank of Canada and major lendersRegulatory watch

Bank of Canada and major lenders reportedly discuss Mythos cyber risk

Bloomberg reported that the Bank of Canada and major banks and financial firms met to discuss cybersecurity risks raised by Mythos.

Canada turns the Mythos question into a cross-border financial-resilience issue rather than a purely U.S. banking story.

View entity profile
Bloombergreported
Medium
Treasury, Federal Reserve, and Major BanksSector watch

Treasury, Fed, and bank CEOs discuss Mythos cyber risk

CBS News reported that Treasury Secretary Scott Bessent and Fed Chair Jerome Powell met with top bank CEOs in a closed-door session about cybersecurity risks posed by Mythos.

Financial-system coordination shows Mythos-class capability being handled as a sector-wide resilience problem, not only as model access.

View entity profile
CBS Newsreported
Critical
AnthropicOfficial

Anthropic announces Project Glasswing and Claude Mythos Preview

Anthropic announced a gated research preview of Claude Mythos Preview through Project Glasswing, naming launch partners and saying access was extended to over 40 additional organizations.

The announcement created the initial access map: named critical-infrastructure partners, unnamed additional organizations, cloud access channels, and explicit defensive-security positioning.

View entity profile
AnthropicofficialAnthropic Frontier Red Teamred-team
Medium
India AI ecosystemStrategic engagement

Anthropic names India a major Claude market through Infosys partnership

Anthropic announced an Infosys collaboration and described India as its second-largest Claude.ai market, with deep technical usage across software and regulated industries.

India is strategically important even without public Mythos access because its developer base and Global South governance posture shape future access debates.

View entity profile
AnthropicofficialAssociated Presswire