Anthropic owned the week, shipping a new frontier model, surfacing a long-rumoured restricted model, and opening a new office — all while OpenAI and Microsoft quietly repositioned for the enterprise battle ahead.
The headline story was the release of Claude Opus 4.8, Anthropic's latest update to its top-tier model. The launch came alongside third-party coverage unpacking what changed, though the most significant backdrop may be Project Glasswing, an internal AI safety initiative that Anthropic published its first public update on Monday. Glasswing appears to be Anthropic's structured attempt to operationalise safety research in parallel with frontier model development — a notable signal that safety work is being formalised as a distinct programme rather than folded into product shipping. The two announcements together — a capable new model and a transparency gesture on safety — reflect Anthropic's ongoing effort to present both axes at once.
The other Anthropic story generating attention was a report suggesting Claude Mythos, a restricted model previously known only to insiders, may be making its way to Claude Code. Mythos has been described as unusually capable at complex reasoning tasks; if it reaches Claude Code users, it would represent a meaningful upgrade for developer-facing workflows. Nothing is confirmed, but the speculation was credible enough to score high.
On the enterprise side, OpenAI published a position piece on what it called the next phase of enterprise AI, and separately outlined its compute infrastructure ambitions. Neither announcement contained product specifics, but together they read as OpenAI framing itself as the default AI infrastructure layer for large organisations. Microsoft reinforced this framing from its own angle, refreshing the Microsoft 365 Copilot design and shipping Copilot Studio updates including improved computer-using agents, a redesigned Workflows experience, and real-time voice capabilities.
Rounding out the week: Meta announced paid AI subscriptions, competing in a tier already occupied by OpenAI and Anthropic, while reportedly facing counter-pressure from OpenAI and xAI eyeing its advertising business. Anthropic also opened its Seoul office, its third in Asia-Pacific, underscoring regional expansion as a priority. And Bloomberg profiled the specialist AI trainers now charging $25,000 a day to help Wall Street firms navigate the agentic shift — a concrete data point on how scarce implementation expertise is at this stage of enterprise adoption.
To watch next week: any developer response to the Claude Mythos-in-Claude-Code reports, and whether OpenAI follows its enterprise positioning piece with concrete product announcements.