This Week's Most Impactful AI News
Special Edition: Anthropic (June 6 – June 12, 2026)
This special edition focuses on Anthropic, which made rapid progress in a week by shipping its best public model, only to be forced offline by a government directive three days later. It increased its valuation lead over OpenAI before a confidential IPO, expanded its security presence across more than 15 countries, and added a renowned researcher to its team. However, the same capabilities that made Anthropic a leader also led to the withdrawal of its top models.
TL;DR: The Anthropic Edition
Claude Fable 5 launched publicly on June 9 as Anthropic’s most capable, widely released model. Three days after release, a June 12 federal export-control directive forced it and its security-focused sibling, Mythos 5, offline for all users.
Anthropic’s $965 billion valuation now sits roughly $113 billion above OpenAI’s, with a confidential IPO filing already submitted and a fall listing possible, though the shutdown adds a fresh regulatory risk to the pitch.
Project Glasswing expanded to about 150 organizations across 15-plus countries after an initial cohort used the model to identify more than 10,000 high- or critical-severity security flaws. But the same June 12 directive knocked its underlying model, Mythos 5, offline mid-deployment.
Andrej Karpathy, OpenAI co-founder and former Tesla AI lead, joined Anthropic’s pre-training team this week, a marquee talent win in a fierce hiring market.
1. Claude Fable 5 Goes Public, Then the Government Pulls It
On June 9, Anthropic launched Claude Fable 5, its most capable available model, and Mythos 5 for vetted partners. They share a core; Mythos 5 has safeguards lifted and boasts strong cybersecurity, having exploited zero-days across major OS and browsers. The public release included guardrails against high-risk requests, with prices dropping to $10 and $50 per million tokens. On June 12, a federal export-control order forced Anthropic to block all foreign nationals’ access; unable to filter users in real time, it disabled both models, while Opus 4.8 remained online. Anthropic challenges the order in court, but the models remain offline, marking what may be the first government-forced takedown of a frontier model.
2. A $965 Billion Valuation and a Confidential IPO
Anthropic now has a $965 billion valuation after closing its Series H round, surpassing OpenAI by about $113 billion, and has filed confidential IPO paperwork with the SEC. Revenue indicates growth: a run rate near $47 billion in May, up from $10 billion a year earlier, with Q2 revenue of around $10.9 billion. Partnerships with AWS, SpaceX, and CoreWeave support growth. A fall listing is possible but unconfirmed. The shutdown complicates the pitch—government orders can instantly restrict flagship models, risking Anthropic’s security edge, which underpins much of its valuation. Underwriters will consider how resilient that advantage is if government control is exercised.
3. Project Glasswing Scales to Critical Infrastructure
On June 2, Anthropic expanded Project Glasswing to around 150 organizations in over fifteen countries, covering power, water, healthcare, communications, and hardware. The program uses Mythos 5 to scan codebases for vulnerabilities; an initial group of fifty partners, with access since April, found over 10,000 high- or critical-severity flaws. However, on June 12, the Mythos 5 tool was pulled offline, cutting off defenders mid-deployment. This same capability, considered too dangerous to operate, was used to secure critical infrastructure, highlighting how fragile this approach is when access can be revoked suddenly.
4. Andrej Karpathy Joins the Pre-Training Team
Andrej Karpathy, an OpenAI co-founder and former head of Tesla’s Autopilot program, started this week at Anthropic on its pre-training team under the lead of Nick Joseph. He is building a group focused on using Claude to accelerate Anthropic’s own training research. The move is a pointed talent win: one of the most respected technical minds in the field, who twice helped build OpenAI, is now improving the model that just overtook it on valuation and capability.
Practical Takeaways
For Individuals:
Don’t anchor your workflow to one model. Fable 5 went from launch to offline in three days; keep a tested fallback, so a single takedown doesn’t stall your work.
Build security review into your AI literacy. Glasswing shows models now find real vulnerabilities at scale; knowing how to direct and verify that work is becoming a marketable skill.
Follow the talent flows, not just the launches. Where researchers like Karpathy move is an early signal of which labs will lead the next cycle.
For Businesses:
Plan for model portability. A government order pulled Anthropic’s two best models overnight; if your product depends on a single model, build and test a fallback before you need it.
Weigh regulatory risk before you standardize. Frontier capability now carries national-security exposure, and the more powerful the model, the more likely access can be restricted.
Treat AI as a security tool, not only a productivity one. Glasswing’s findings suggest frontier models belong in your defensive stack, with human verification.
Conclusion & Footer
Anthropic had a busy week, releasing a frontier model and a stronger restricted version, gaining a valuation lead over OpenAI, expanding security into critical infrastructure, and hiring a top researcher. Soon after, a federal order forced its best models offline. The same capability that gave it an edge also attracted government attention. The lead is real, but so are the risks: as models improve, authorities may restrict access.