Anthropic Forced to Pull Fable 5 and Mythos 5 After US Export Control Order The Verge / TechCrunch The Trump administration issued an export-control directive on Friday ordering Anthropic to suspend all access to its flagship Fable 5 and Mythos 5 models for foreign nationals. Anthropic initially believed the order was triggered by a jailbreak of Fable 5’s cybersecurity capabilities, but TechCrunch’s reporting suggests the real motivation may be political — driven by “personality differences” between Anthropic leadership and the administration — rather than any specific technical threat. The forced shutdown has reignited debate about government reach into the AI industry and the risks of building products on models your company doesn’t own.

🏛️ Policy & Safety

Inside the fight over Claude Mythos 5 — The Verge At 5:21 PM on Friday, Anthropic received a US export control directive to suspend access to Mythos 5 and Fable 5 — delivered mid-World Cup celebration weekend. The full narrative of how Anthropic fought the directive before ultimately complying sheds light on just how unprepared the AI industry was for this kind of government intervention.

The US government’s Anthropic models ban was never about an AI jailbreak — TechCrunch TechCrunch’s Zack Whittaker argues the directive was reactive and possibly retaliatory, concluding that the AI industry is no longer insulated from the kind of government interference that has long shaped biotech and defense. The piece underscores how much regulatory risk now sits inside frontier-model companies.

Anthropic’s Safety Superpower — Stratechery Ben Thompson’s analysis of the Mythos situation argues Anthropic’s cautious rollout was genuinely justified — Mythos 5 is materially more capable at identifying and exploiting security issues than any prior generation, and guardrail jailbreaks surfaced quickly after release. The piece defends Anthropic’s positioning while acknowledging the model’s power demands extraordinary care.

The Once And Future Fable #2 — Zvi Mowshowitz / Substack Zvi’s detailed breakdown of the policy fallout notes all the unknowns: what motivated the directive, what fix is being demanded, and what the administration intends next. He treats it charitably as a potential misunderstanding that could be sorted quickly, while noting the information vacuum is itself damaging.

UK bans under-16s from using social media apps including TikTok and YouTube — AP News — The ban is set to take effect early next year, making the UK one of the first major democracies to legislate hard age gates on social platforms.

DOJ claims xAI’s unpermitted gas turbines are a matter of ’national, economic, and energy security’ — TechCrunch — The Justice Department has intervened on xAI’s behalf, arguing that the Pentagon requires the company’s Memphis data center to stay operational.

🚀 Industry Moves

SpaceX is officially buying Cursor for $60 billion — The Verge Days after its blockbuster IPO, SpaceX announced it will acquire AI coding assistant Cursor in an all-stock deal valued at $60 billion — one of the largest acquisitions in tech history. The move is designed to accelerate SpaceX’s struggling AI division and win enterprise developer customers away from competitors like Anthropic and OpenAI. Cursor’s team and product will remain intact, at least initially.

SpaceX valuation balloons to $2.6T, briefly passes Amazon — TechCrunch SpaceX’s public market debut added more than $1 trillion in market cap in 48 hours, briefly vaulting it ahead of Amazon. The company has positioned itself as simultaneously a launch provider, an internet company (Starlink), and now — through the Cursor acquisition — an AI platform.

ChatGPT’s market share slips below 50% for first time — TechCrunch OpenAI’s flagship product still leads with over 1.1 billion monthly users, but it has fallen below 50% market share for the first time as Gemini (662M users) and Claude (245M) eat into its dominance. The milestone marks the AI assistant market’s shift from a near-monopoly to a genuine multi-player race.

Fox to Buy Roku Streaming Service in $25 Billion Deal — WSJ — The acquisition will combine Fox’s subscription streaming services with Roku’s hardware and ad platform, positioning the pair to compete directly with Amazon and Netflix for ad-supported video revenue.

Owning vs. Renting Intelligence — Lin Qiao / Fireworks CEO — Fireworks CEO used the Mythos shutdown to argue that companies whose core product runs on models they don’t own are exposed to risks they can’t control, and that post-training open models is an increasingly rational hedge.

🔬 Frontier Research

Google DeepMind Explores the Path to ASI — Google DeepMind A new DeepMind paper outlines four possible pathways from AGI to artificial superintelligence, examines likely bottlenecks (data, compute, algorithmic efficiency, alignment), and addresses the societal implications of each trajectory. It’s one of the first formal published frameworks for thinking about post-AGI development from a major lab.

DFlash and Spec V2 Decoding — LMSYS — The latest generation of speculative decoding with DFlash and SGLang’s Spec V2 engine shows substantial throughput gains over baseline inference; a significant practical result for inference-heavy deployments.

Accelerating multilingual AI with a new open dataset — GitHub Blog — GitHub released a repository-level metadata dataset surfacing public repos with non-English natural-language content, designed to close the gap in multilingual model training data.

🛠️ Developer Tools

Android 17 launches with new multitasking tools as Google expands Gemini features — TechCrunch Google shipped Android 17 alongside Wear OS 7 and a Pixel Drop, with Gemini integration running deeper into the OS than any prior release. The update includes new parental controls, security tooling, and multitasking improvements designed to take advantage of foldable and large-screen form factors.

Agentic Code Review — Addy Osmani A widely-shared data-driven piece arguing that coding agents have moved the hard part of engineering from writing code to deciding whether to trust it. Faros AI’s study of 22,000 developers found code churn up 861% and defect rates up from 9% to 54% since AI coding tools went mainstream — while GitClear data shows 4x raw output for only ~12% delivered-value gain. Code review is now the highest-leverage engineering skill.

Google Chrome’s next update will mark the end of popular ad blockers — 9to5Google — Chrome 151 removes all Manifest V2 support, ending years of delays and finally killing the extension API that most major ad blockers depend on.

Apple Foundation Models — Claude SDK for Swift — Anthropic — A new Claude for Foundation Models Swift package lets developers call Claude through Apple’s Foundation Models framework on Apple platforms.

Claude Code 2.1.178 — Anthropic — This release adds Tool(param:value) syntax for permission rules (e.g., Agent(model:opus) to block Opus subagents) and improves nested .claude/ directory skill loading.

💼 Enterprise & Startups

Facebook’s new AI Mode search gets its info from public posts — The Verge — Meta’s new conversational search mode mines public Facebook Groups, Reels, and Marketplace listings; critics have raised accuracy and privacy concerns about crowd-sourced AI summarization.

Robinhood’s note on 10% layoffs shows blaming AI isn’t cutting it — TechCrunch — CEO Vlad Tenev’s layoff announcement made no mention of AI, standing out from an industry that has normalized AI-as-pretext for headcount cuts.

Plaud says its software business topped $100M in ARR after shipping over 2M AI notetakers — TechCrunch

Probably raises $9M to build a more reliable kind of AI — TechCrunch — Probabilistic AI startup targeting hallucination prevention and deterministic-level accuracy.

A backdoor in a LinkedIn job offer — roman.pt — A developer was asked to review a public GitHub repo falsely attributed to a real engineer — a social-engineering vector worth knowing about as AI-assisted recruiting scales.

Generated by claude-sonnet-4-6 · 2026-06-16T10:00:00Z