Anthropic Weekly: Claude Tag ships, Mythos partially returns, and Alibaba claims widen the risk story

Anthropic's week split cleanly in two: product momentum on the inside, access-risk turbulence on the outside. Claude Tag turns Slack into a shared workspace for delegated Claude tasks; Mythos 5 moved from a global shutdown to a limited U.S. critical-infrastructure redeployment; and the export-control fight is now feeding both copycat-risk claims and regional AI alternatives.

Anthropic launched Claude Tag on June 23 as a Slack-based beta for Claude Enterprise and Team customers. Administrators can grant Claude access to selected channels, tools, data, and codebases; users can then tag @Claude in a Slack channel and delegate work into a thread. Anthropic says the product can remember relevant channel context, schedule work for later, and run asynchronously over hours or days. The company also says its internal version now creates 65% of its product team's code, and that Claude Tag replaces the prior Claude in Slack app with a 30-day administrator migration window. 1

The important shift is not the Slack integration by itself. Claude Tag gives teams a scoped agent identity: a sales-channel Claude should not pass memories to an engineering-channel Claude, and administrators get spending limits plus logs of what Claude did and who requested it. That makes the product a direct test of whether Anthropic can sell enterprise agents without making security teams feel they have invited an unbounded bot into every system. 1

The Fable 5/Mythos 5 access story did not resolve, but it moved. Reuters reported that Mythos had identified vulnerabilities in highly sensitive U.S. government systems during testing under Project Glasswing. Senator Mark Warner said in a hearing that he had been told by NSA chief Joshua Rudd that Mythos "broke into almost all of our classified systems, not in weeks, but in hours"; Reuters also noted the AP's caveat that identifying vulnerabilities within hours did not mean the model exploited them within that time. 2

Two days later, Reuters reported that the U.S. government allowed Anthropic to release Mythos 5 to some "trusted" U.S. organizations, partially reversing the June 12 order that had suspended access over national-security risks. A source told Reuters that more than 100 companies and institutions would get access, including many Fortune 500 companies; Anthropic's own statement said Mythos 5 could be redeployed to U.S. organizations that operate and defend critical infrastructure. 3

Fable 5 remains the unresolved half. Reuters separately reported, citing Axios, that the administration was close to allowing Anthropic to restore Fable 5, while Reuters itself could not immediately confirm the report. That should be treated as a watch item, not a completed reopening. 4

Reuters reported on June 25 that Anthropic accused Alibaba of illicitly extracting Claude's model capabilities, describing it as the largest known attack of its kind on the company, according to a letter seen by Reuters. 5

That report matters because it changes the frame around frontier model restrictions. The U.S. order was already a customer-access problem. The Alibaba allegation makes it a defensibility problem: if advanced capabilities can be copied or approximated through adversarial use, Anthropic has to convince customers, regulators, and future public-market investors that access controls are more than compliance theater. The claim is still an allegation, not a court finding.

The shutdown's market effects are no longer hypothetical. Reuters reported that Chinese startup Z.ai said it plans to use domestic-listing proceeds to fund its AGI work after its GLM-5.2 model scored close to leading U.S. models from Anthropic and OpenAI on public benchmarks. Reuters also reported that GLM-5.2 held fourth place on Artificial Analysis' LLM intelligence leaderboard and second on Code Arena's front-end coding leaderboard, while operating at roughly one-sixth the cost of closed U.S. frontier models. 6

TechCrunch tracked the same pressure in Asia. Sakana AI launched Fugu, which it positioned as standing alongside Anthropic's Fable 5 and Mythos Preview, while Chinese cybersecurity firm 360 reportedly unveiled Tulongfeng, an AI tool it said could compete with Mythos. Sakana told TechCrunch the timing was coincidental, but its marketing also leaned into the export-control moment by promising frontier capability without export-control risk. 7

For Anthropic, the risk is not only lost revenue in Asia. If local models become the default during a U.S. access freeze, switching costs start to work against U.S. labs even if restrictions later loosen.

Anthropic joined RAISE US as a founding partner, according to its official X post. Fortune, carrying AP reporting, said the nonprofit is starting with more than $500 million for education and training programs, with Amazon, Microsoft, Anthropic, the OpenAI Foundation, and Bank of America among its anchor partners. 8 9

Anthropic also published a new Economic Index report. The report says personal Claude conversations rise from about 35% on weekdays to just under 50% on weekends; tax-related requests were eight times as common on April 14 as on an average day in May; and its linked survey sample had about 9,700 respondents after filtering. More than one-third of respondents expected AI to be able to do most or nearly all of their work tasks within 12 months, while 10% rated losing their own job as likely or very likely. 10

On leadership, Reuters reporting carried by The Economic Times said Steve Jarrett, Orange's chief AI officer, left the French telecom group to join Anthropic. The report said he will start on August 25, be based in Paris, and initially help Anthropic adapt products for European and African markets. 11

Claude Tag is the cleanest product signal: Anthropic is pushing Claude from individual assistant to organization-level agent, with identity, permissions, logs, and spend controls at the center. That is exactly where enterprise buying committees will pressure-test the product.

The harder story is access. Mythos 5 now has a narrow path back for U.S. critical infrastructure, but the wider Fable 5 reopening remains uncertain. Meanwhile, the Alibaba allegation and the Z.ai/Sakana/360 responses show that restricted access can create a vacuum. Anthropic's next few weeks will hinge on whether it can restore broader availability without losing the policy argument that made the restrictions politically salient in the first place.