Anthropic disables Fable 5 and Mythos 5 worldwideThe Verge / Bloomberg / Axios The Trump administration issued an export-control directive ordering Anthropic to block all foreign access to its frontier models Fable 5 and Mythos 5, citing national security concerns after a third party claimed to have jailbroken Mythos. To comply, Anthropic had no choice but to cut off all customers globally — including its own employees. The company pushed back publicly, writing: “We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.”

🔒 Policy & Security

Anthropic cuts off Fable 5 and Mythos 5 access following government orderThe Verge

Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei declaring Fable 5 and Mythos 5 subject to export controls covering any location outside the US and all foreign persons within it. With no technical mechanism to restrict access only to US citizens, Anthropic abruptly disabled the models for every customer worldwide. The company warned the precedent could effectively freeze all frontier model deployments across the industry if other labs face the same standard.

OpenAI faces investigation from state attorneys generalTechCrunch

A coalition of 42 state AGs — led by New York — served OpenAI with a sweeping subpoena covering its advertising practices, user engagement and retention tactics, model sycophancy, handling of health data, and treatment of minors and seniors. The probe follows a separate Florida AG lawsuit earlier this month alleging ChatGPT is an unsafe product. OpenAI says it is cooperating “constructively.”

Chinese cybercrime operation that used AI to scam ‘hundreds of thousands of victims’ sued by GoogleTechCrunch — Google sued the group “Outsider Enterprise” after it used AI tools to blast 2.5 million fraudulent text messages over a two-week span, targeting hundreds of thousands of victims.

🤖 Frontier Models

Anthropic’s safety warnings may have just backfiredTechCrunch

A pointed analysis of the political irony: Anthropic’s years of public messaging about frontier-model risks appears to have primed the government to act decisively — and against Anthropic itself — the moment a jailbreak report surfaced. The piece explores how the company’s own safety framing created the regulatory scaffolding used to shut it down.

GPT-5.6 “Kindle-Alpha” leak circulates in developer channelsWindows Forum / AI News Today — A checkpoint labeled GPT-5.6 “kindle-alpha” reportedly surfaced through a misconfigured Codex API endpoint. Community reports describe stronger reasoning and a 1.5M-token context window; OpenAI has made no official comment.

Ramp AI Index: Anthropic overtakes OpenAI in US business adoption for the first timeBuild Fast With AI — The latest Ramp spending data shows Claude edging out ChatGPT in enterprise seat share, even as a separate IDC survey paints a more cautious picture of Claude’s broader reach.

🏢 Enterprise & Industry

Meta’s months-old AI unit is a soul-crushing gulag, say the engineers stuck inside itTechCrunch

A new report finds Meta’s 6,500-person AI division — formed through a massive internal consolidation earlier this year — on the verge of revolt. Engineers describe bureaucratic gridlock, unclear mandates, and a culture mismatch between Meta’s fast-ship ethos and the long-horizon research needed to compete with OpenAI and Anthropic.

The future of Hollywood isn’t feeding prompts into vanilla gen AI modelsThe Verge — Tribeca 2026 coverage notes that most AI video models can still only produce short clips unsuitable for theatrical release. Filmmakers working with Google DeepMind and OpenAI are building proprietary fine-tuned pipelines rather than using off-the-shelf models.

🛠️ Developer Tools

Kiro CLI 2.7: /goal, Queue Steering, and enriched /rewindKiro — Queue steering lets users redirect the agent mid-task at tool boundaries; /goal creates iterative loops that verify completion before stopping; /rewind now shows a detailed turn-by-turn picker.

Claude Code 2.1.176Claude Code — Session titles now generate in the conversation’s language; adds footerLinksRegexes setting for customizable link badges; improved Bedrock credential caching tied to actual token expiration.

📱 Consumer AI

Apple’s new AI photo editing tools mostly work, for better and worseThe Verge — A hands-on with iOS 27’s new Reframe, Extend, and Clean Up features. More conservative than Pixel’s Magic Eraser family, but integrated tightly enough that casual users will find them useful for the first time.

My yard is dying, so I made an app for thatThe Verge — A consumer vibe-coding chronicle: the author used Gemini to build a functioning backyard management app in an afternoon, bugs included.


Generated by claude-sonnet-4-6 on 2026-06-13T10:00:00Z