SK Telecom Loses Claude Mythos as AI Access Fight Spreads

The SK Telecom Claude Mythos dispute signals that Washington is starting to treat access to frontier AI models less like a software subscription and more like a controlled strategic asset.

That is the real story beneath the shutdown. Days before Anthropic took Claude Mythos and Fable 5 offline, the White House ordered the company to revoke SK Telecom’s access to Claude Mythos over alleged China links, according to Wired. Anthropic complied, but the episode did not end there. The White House later ordered access to Mythos and Fable 5 restricted to US nationals, including by cutting off foreign nationals inside the US. Anthropic chose to disable the models entirely rather than build nationality-based access controls that would be hard to implement while preserving privacy.

That move follows the escalation we covered in White House Forces Anthropic Fable Shutdown in AI Feud and Anthropic Export Controls Throw AI Access Into Chaos. The new detail is sharper: a major South Korean telecom carrier became the flashpoint.

SK Telecom Claude Mythos access became a test of who gets frontier AI

XOOMAR analysis: the White House’s concern was not just one company. It was control. Anthropic had opened early Claude Mythos access through Project Glasswing, a program for trusted organizations using the model to identify software vulnerabilities. Earlier this month, SK Telecom became one of roughly 150 companies to receive access as Anthropic expanded the program “following several weeks of close collaboration” with outside experts and the US government.

The model matters because Wired describes Claude Mythos as “exceptionally skilled at identifying software vulnerabilities.” Fable 5, released publicly on June 9, was a highly safeguarded version of Mythos. Amazon researchers later told the White House they found ways to bypass some Fable 5 guardrails and reach Mythos’ cybercapabilities. Anthropic and outside cybersecurity experts disputed the idea that those risks were unique to Claude.

The counterpoint is important. A person close to Anthropic told Wired the company viewed SK Telecom’s Mythos access and Amazon’s Fable 5 findings as separate issues. The government’s letter demanding restrictions to US nationals also did not reference the Korean company or China.

Still, the sequence matters. SK Telecom’s access was revoked first. Then the broader shutdown came. That makes the SK Telecom Claude Mythos case a live example of how geopolitical doubts can collide with technical safety concerns and trigger a platform-level cutoff.

The hard numbers make the alleged China link narrower, not irrelevant

The available data does not show SK Telecom as a China-heavy operator, but it does show a history Washington could scrutinize. Wired reports that SK Telecom generated only about $1.9 million in revenue from China in 2024, mainly from investment-related activities, and employed just seven people there, based on its annual report.

The larger issue is SK Telecom’s corporate and historical footprint. The company is part of SK Group, whose affiliates have business interests in China across semiconductors, energy, and other industries. SK Telecom also had a long-running connection to China Unicom.

The strongest counterpoint comes from SK Telecom itself. In response to earlier Washington Post reporting that did not name the company, SK Telecom told a Korean newspaper:

“anonymous insider’s remarks in foreign media lack verified facts, and our company has no ties to China.”

That denial matters. The public record Wired lays out does not prove that SK Telecom gave China access to Claude Mythos, nor does it establish operational control by a Chinese entity. But Washington often acts on perceived pathways, not courtroom-ready proof. In this case, the pathway appears to be historical telecom ties, group-level China exposure, and access to a model with advanced cyber capabilities.

Anthropic, SK Telecom, Washington, and Seoul are playing different games

XOOMAR analysis: each side has a rational position, and those positions clash. SK Telecom wants access to advanced AI because it has already invested in Anthropic and partnered with the company to develop an AI model tailored to the telecommunications industry. Losing access to Claude Mythos cuts against that strategy and puts the company in the awkward position of defending its credibility without seeing all the evidence behind Washington’s concerns.

Anthropic’s problem is more direct. The company markets itself around AI safety, yet Wired reports that the White House concluded it could not trust Anthropic to safeguard its most advanced AI technology after the SK Telecom dispute and Amazon’s Fable 5 findings. Anthropic declined to comment to Wired. That silence may be legally prudent, but it leaves customers guessing about the rules for future access.

Washington’s incentive is containment. The administration ordered access restricted to US nationals, a blunt standard that reaches foreign nationals inside the US. That is a much broader move than revoking one foreign customer.

Seoul’s concern is also clear from the facts available. South Korean organizations in Project Glasswing included Samsung Electronics and the Korea Internet & Security Agency, alongside SK Telecom. If one national champion is treated as a risk because of alleged China ties, other Korean institutions will want clearer standards before building sensitive workflows around US frontier models.

The Mythos fight extends the old China tech playbook into model access

This dispute follows a familiar US-China pattern, but the asset is different. Wired notes that in 2021, the first Trump administration restricted US investment in China Unicom as part of a broader effort targeting Chinese firms Washington said were linked to China’s military and intelligence sectors. In April of this year, the Federal Communications Commission proposed barring US telecom firms from interconnecting with China Unicom and other Chinese carriers, citing national security concerns.

Claude Mythos is not a carrier network or a semiconductor shipment. It is model access. That makes the control point more fluid. A user can be added to a program, removed from it, or blocked at the account layer. Anthropic’s decision to take the models offline rather than filter by nationality shows how messy that becomes when privacy, identity, and enterprise access all collide.

The counterpoint is that Anthropic did not appear to fight the initial SK Telecom revocation. Wired reports the company immediately complied, and the government did not threaten export controls on the model at that time.

The thesis still holds because the later order expanded from a single customer decision to a nationality-based access regime. That is the leap. The government did not just question a customer. It reshaped the distribution model.

Enterprise AI buyers now have a new diligence problem

The SK Telecom Claude Mythos case turns geopolitical exposure into an enterprise AI availability risk. Buyers already compare model performance, security posture, and integration costs. After this episode, they have to ask a more uncomfortable question: can access disappear because a government dislikes their ownership structure, old joint ventures, group affiliates, or foreign partnerships?

The immediate operational risk is simple. A team that builds workflows around a frontier model can be stranded if access is revoked with little notice. In this case, Anthropic did not merely remove one company. It disabled access to Mythos and Fable 5 entirely after the White House order.

Contracts will likely need sharper language around government orders, termination rights, customer disclosures, audit rights, and who counts as an authorized user. That is XOOMAR analysis, but it follows directly from the practical problem Anthropic faced: nationality-based gating was difficult enough that the company chose a full shutdown instead.

What would weaken this thesis? If Anthropic and the White House bring Mythos and Fable 5 back under narrow, transparent rules that do not disrupt approved allied customers, this may look like an emergency misfire. If access returns only through case-by-case approvals, the strategic asset model wins.

The next fights will turn on allies, audits, and model kill switches

The next AI access battles will not be limited to China. They will run through allies whose companies have any China-adjacent exposure Washington finds risky. That is the uncomfortable lesson for South Korea. SK Telecom is based in a US-aligned democracy, yet it became central to an export-control fight over Anthropic’s most powerful AI technology.

Frontier AI labs will face pressure to build stronger customer screening, usage monitoring, and emergency shutoff systems for high-capability models. Governments will want faster intervention tools. Customers will want due process and predictable standards.

The evidence to watch is concrete: whether SK Telecom regains access, whether Anthropic publishes clearer criteria for Project Glasswing, whether US officials create formal licensing rules for foreign access, and whether allied institutions such as Samsung Electronics or KISA face similar scrutiny. If those controls harden, the Mythos episode will become the template: frontier models distributed less like software, and more like strategic infrastructure with a kill switch.

Impact Analysis

• The dispute shows frontier AI access is increasingly being treated as a national security issue.
• SK Telecom’s role highlights how allied foreign companies can still face scrutiny under US AI controls.
• Anthropic’s shutdown underscores the technical and privacy challenges of enforcing nationality-based AI restrictions.