AI Business | Jun 29, 2026

Analysis

• Chinese labs closed the cybersecurity-capability gap in the same week the US began reopening access to the models it had restricted: Zhipu AI's open-weight GLM-5.2 matched Anthropic's Mythos on bug-finding while the Commerce Department let Anthropic restore Mythos 5 access and moved to clear Fable 5's return after a four-month freeze — a freely downloadable Chinese equivalent complicates the rationale for the US restrictions.
• The frontier of cybersecurity capability is now split by distribution model: Anthropic's Mythos stays gated behind access approved by the US government, while the Chinese model claiming parity, Z.ai's GLM-5.2, is open-weight and runnable by anyone on standard hardware — who can deploy the capability no longer tracks which model is nominally ahead.
• Compute capital kept moving into new geographies as the model race tightened: Australia's Firmus will build its first Indonesia data center with Nvidia, anchoring as much as $30 billion in committed offtake — Nvidia securing Southeast Asian capacity even as Washington re-litigates who may access the frontier models that capacity is built to serve.

POLICY & REGULATION

• TRADEAxios | Powerful Anthropic model Fable 5 on track to return soon | Axios reported the Commerce Department on Friday allowed Anthropic to restore access to Mythos 5, its strongest cybersecurity model, for a limited number of trusted users — a thaw in a four-month standoff after the administration pulled Anthropic's Fable 5 offline through government intervention. Commerce Secretary Howard Lutnick wrote that Anthropic "has worked with the U.S. government to address risks associated with" the models and had committed to joint protocols and release standards. Fable 5's return still requires sign-off from the Pentagon and National Security Agency; Axios reported Treasury Secretary Scott Bessent and Lutnick helped defuse the fight.

RESEARCH & MODELS

• MODELThe Wall Street Journal | China Has Matched Anthropic in Cybersecurity, Resetting AI Race | The Wall Street Journal reported Chinese AI systems have matched Anthropic's powerful model Mythos in some cybersecurity bug-finding scenarios, a development it said resets the US-China model race and is pressuring the White House on AI policy. The Verge reported the model behind the parity claim is Zhipu AI (Z.ai)'s newly released open-weight GLM-5.2, which still lags Anthropic and OpenAI models on more general tasks but has sharply narrowed the capability gap. Because GLM-5.2 is open-weight, it can be downloaded and run on readily available hardware by anyone, including actors outside US oversight.

COMPUTING & INFRASTRUCTURE

• DCBloomberg | AI Startup Firmus to Build Indonesia Data Center With Nvidia | Bloomberg reported Australian AI-infrastructure firm Firmus Technologies will build its first Indonesia data center in partnership with Nvidia, a project the company said is expected to win as much as $30 billion in committed offtake agreements.

MARKETS

25 Jun 2026 close | Retrieved 25 Jun 22:00 UTC | Yahoo Finance

AI Equities (1D) | Nvidia 195.74 USD -1.6% | Microsoft 352.83 USD -3.5% | Alphabet 343.71 USD -0.5% | Meta 543 USD -2.7% | Amazon 227.01 USD -3.1% | Palantir 107.27 USD -5.5%

Semiconductors (1D) | AMD 533 USD +2.5% | TSMC 434.99 USD -1.3% | Broadcom 378.91 USD -0.8% | ARM 347.71 USD -3.2% | Super Micro 31.68 USD -2.4%

AI Infrastructure (1D) | CoreWeave 98.76 USD -2.1%

Indices (1D) | NASDAQ 25,359 -0.5% | SOX 13,941 +3.6%

Coverage: 27 Jun 01:00 – 29 Jun 01:00 UTC