JPMorgan Cuts Anthropic Access in Hong Kong, Citing Licensing Terms

JPMorgan Chase has stopped its Hong Kong staff from accessing Anthropic’s AI models, according to a Financial Times report on Thursday that cited three people familiar with the matter, a move that lifts a curtain on how Wall Street’s biggest bank is reading its contract with one of the most consequential AI labs in finance. The trigger, the FT reported, was the wording inside Anthropic’s licensing agreement with the bank. Goldman Sachs had taken the same step in April.

JPMorgan staff in Hong Kong can no longer pull Anthropic’s Claude models from the bank’s internal drop-down of approved large language models, the FT said. JPMorgan and Anthropic did not respond to requests for comment outside business hours, according to the the Financial Times report on JPMorgan’s Hong Kong move.

The Hong Kong Cut-Off

Inside JPMorgan’s Hong Kong offices, the change is mechanical: Claude disappears from the approved tools list, the same list that still surfaces ChatGPT, Gemini and other sanctioned models. Anthropic has long taken the position that Hong Kong is not part of the markets where its API and Claude.ai are officially accessible. A spokesperson for the lab told the FT, in coverage of Goldman’s earlier move, that its Claude models had never been officially supported in Hong Kong, a posture that put the burden on each bank to police its own usage.

Western AI products are unavailable in mainland China under the Great Firewall, but Hong Kong has largely remained outside that filter, with usage limits set by US providers rather than Beijing. That gray zone is what Goldman Sachs and now JPMorgan have decided is too narrow to keep Claude on the menu. Reuters reported earlier in the week, via an the Reuters-sourced account of JPMorgan’s licensing trigger, that the move was driven by the language of Anthropic’s usage terms inside the JPMorgan licensing deal.

Why Anthropic’s Licensing Forced JPMorgan’s Hand

The clause that pushed Claude off the Hong Kong list is the same clause Anthropic points to whenever pressed on its geographic footprint. Anthropic’s website does not list Hong Kong as a market where its API and Claude.ai are officially accessible, a stance the lab reiterated to the FT during the Goldman coverage earlier this year. JPMorgan, like Goldman before it, opted to read that wording conservatively.

The cut-off also lands in a week when Anthropic is under direct US government pressure to wall off its frontier models from foreign users. Earlier this week, US Commerce Secretary Howard Lutnick, in a letter to Anthropic chief executive Dario Amodei, ordered the company to suspend exports of its Mythos and Fable AI models to destinations worldwide and to all foreign nationals, citing concerns they could be used by military intelligence users in China, Russia and other countries of concern. Anthropic said on June 13 it will “abruptly disable” Fable 5 and Mythos 5 for all customers to comply, while arguing it had been given only “verbal evidence of a potential narrow, non-universal jailbreak.” President Donald Trump said Wednesday that negotiations with Anthropic are “going fine.”

Inside JPMorgan’s compliance shop, the mix is hard to separate. Anthropic’s own licensing terms exclude Hong Kong, the Commerce Department is pushing the lab to keep foreign nationals off its strongest models, and US-China tensions over AI technology, data security and access to advanced computing tools are rising. Each of these threads pushed the same direction: remove Claude from the Hong Kong drop-down. Some American AI companies fear “distillation,” where local actors train new models through use of foreign ones, and Anthropic’s wider geographic squeeze fits that fear.

• April 7, 2026: Anthropic commits up to $100M in usage credits for Mythos Preview across Project Glasswing, plus $4M in direct donations to open-source security organizations.
• June 13, 2026: Anthropic says it must disable Fable 5 and Mythos 5 for all customers after a US export control directive.
• June 17, 2026: Trump tells reporters negotiations with Anthropic are “going fine.”

The Goldman Precedent From April

JPMorgan is not the first US bank to act. In April, Goldman Sachs removed Claude from a list of approved tools available to its Hong Kong-based bankers, according to the FT. The bank had consulted with Anthropic and concluded, after a strict interpretation of the contract, that staff in Hong Kong should not be allowed to use any of the lab’s products. Employees had been unable to access Claude models directly or through Goldman’s internal AI platform for several weeks before the change was widely noticed.

The Goldman cut did not extend to other AI providers. Hong Kong staff at Goldman retained access to ChatGPT and Gemini, a carve-out that points to where the boundary actually sits. Anthropic’s licensing geography is the binding constraint, not a blanket ban on US-built models in Hong Kong. Some US AI companies fear “distillation,” where local actors train new models through use of foreign ones, and Microsoft and OpenAI began investigating in January whether a group linked to Chinese AI startup DeepSeek improperly obtained data from OpenAI’s technology.

JPMorgan’s Bigger Anthropic Bet

The Hong Kong removal sits next to the most aggressive Anthropic partnership on Wall Street. On April 7, 2026, Anthropic launched Project Glasswing, an initiative that brings together Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA and Palo Alto Networks in an effort to secure the world’s most critical software. JPMorgan’s chief information security officer, Pat Opet, signed on as a launch partner, and Anthropic committed up to $100M in usage credits for Mythos Preview across the project. Read the Anthropic’s Project Glasswing announcement and partner list for the partner roster and the Pat Opet statement.

JPMorgan is the biggest US bank by assets and runs a nearly $20 billion annual technology budget. In a June 9 interview with CNBC, the bank’s chief analytics officer, Derek Waldron, said JPMorgan plans to deploy artificial intelligence agents later this year that can work autonomously for far longer than existing versions, “an hour or two” rather than “two or three minutes,” a class of long-running agents that Waldron said the firm will have in 2026. The bank has seen a 20% increase in gross sales because of these tools, Waldron said, and believes they could eventually let individual bankers expand client coverage by as much as 50%. Waldron also said long-running agents are not yet ready for corporate use due to security concerns.

Anthropic’s wider footprint inside the bank, the IPO-bound lab’s other recent context and the Mythos rollout sit on the other side of the licensing wall JPMorgan just enforced in Hong Kong. Read Anthropic’s $965 billion valuation and the Mythos rollout for the broader stake JPMorgan is taking.

The Cybersecurity Risk JPMorgan Is Bragging About

It’s a “very heightened risk.”

That is how Jamie Dimon, chief executive of JPMorgan Chase, described the cybersecurity threat from Anthropic’s Mythos at an Anthropic event in New York, according to American Banker. Read Dimon’s “very heightened risk” remarks at the Anthropic event for the full panel. Dimon noted on the bank’s earnings call that Mythos “does create additional vulnerabilities” and that AI has made cyber risk worse, even as he said JPMorgan is “very well protected.” He wrote in his annual shareholder letter that cybersecurity has been the bank’s biggest risk “for years,” two weeks before Anthropic announced Mythos.

Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell called an emergency meeting with big-bank CEOs about Mythos shortly after its restricted April 7 release, according to Banking Dive. Fed Vice Chair for Supervision Michelle Bowman said at a Financial Stability Oversight Council roundtable that Mythos “highlights the dynamic nature” of AI and “if used maliciously it could be deployed to identify and exploit weaknesses,” and that the Fed will consider how to effectively supervise its use. The IMF, separately, has warned that AI-driven cyberattacks could trigger a global financial shock. Read the IMF warning on AI-driven financial shocks.

Anthropic has been public about what Mythos Preview has found inside critical software, and the disclosures are concrete:

1. A 27-year-old vulnerability in OpenBSD that let an attacker remotely crash any machine running the operating system by connecting to it.
2. A 16-year-old vulnerability in FFmpeg, the video codec used across countless pieces of software, in a line of code that automated testing tools had hit five million times without catching the problem.
3. A chain of vulnerabilities in the Linux kernel that let an attacker escalate from ordinary user access to complete control of the machine.

Anthropic said that with Mythos it found almost 300 vulnerabilities in Firefox, compared with about 30 with the model that preceded Mythos. JPMorgan is betting that helping Anthropic find and patch those flaws is the surest defense for the bank’s own systems. In Hong Kong, the same lab’s models are off the menu for a different reason. Anthropic said it is working to restore access to Fable 5 and Mythos 5; until then, the bank’s Hong Kong bankers will not see Claude in their drop-down, while their New York colleagues press deeper into the same lab’s frontier tools.