AI News Roundup — June 14, 2026
The full picture of Anthropic's Fable 5 shutdown came into focus: the UK's AI Safety Institute jailbroke the model within hours of access, Amazon's Andy Jassy personally called senior US officials after his researchers extracted cyberattack-usable information, and Dario Amodei reportedly refused to patch the flaw — triggering a global firestorm, European geopolitical backlash, and serious questions about Anthropic's IPO timeline.
UK Safety Institute Jailbroke Fable 5 Within Hours — The Full Shutdown Timeline Revealed
New details from the Wall Street Journal and SiliconAngle clarify the chain of events behind Anthropic's forced shutdown. On June 9 — the same day Fable 5 launched — Xander Davies, Red Team lead at the UK's AI Safety Institute, posted on X: "Within a few hours of access, we developed a jailbreak that extracted malicious responses to single-turn question-answering." He added that over two additional days of testing, his team extended the jailbreak to "sometimes allow for multiple steps of malicious agentic tool-calls." Separately, Amazon's own researchers used Fable 5 to obtain "information that could be used in cyberattacks" and CEO Andy Jassy personally called Treasury Secretary Scott Bessent and other senior officials to raise the alarm. The White House asked Anthropic CEO Dario Amodei to patch or de-deploy Fable 5. Amodei declined, calling it a "minor" risk also present in GPT-5.5 and other rival models. Commerce Secretary Howard Lutnick then issued the export control directive on Friday evening. Anthropic, unable to screen by nationality at API level, disabled Fable 5 and Mythos 5 globally. The DoD's Chief Information Officer posted: "Some things are more important than revenue cycles, clickbait, and pre-IPO valuation. America First. Always."
SiliconAngle · Wall Street Journal · Al Jazeera
Europe Erupts — France and UK Call Shutdown a 'Wake-Up Call' for AI Sovereignty
The global Fable 5 shutdown ignited a geopolitical firestorm across US allies. British MP Tom Tugendhat said dozens of UK hospitals and research institutions lost access to the model overnight, some mid-study. "Disabling Fable 5 for foreigners is not a misunderstanding or a mistake — it's the inevitable result of technology shaping warfare so that sovereignty is more about code than cannons," he said. French politicians Jordan Bardella and Bruno Retailleau called the ban a "wake-up call" for Europe and urged investment in domestic AI companies — specifically naming Paris-based Mistral AI — to ensure critical digital infrastructure cannot be "switched off on a whim by Washington officials." The European Commission said it is "assessing the practical consequences" and warned that contingency measures "should not be discriminatory against partners." India's Zoho co-founder Sridhar Vembu: "Technology is the ultimate weapon. National sovereignty is now about technology." The episode is accelerating momentum for AI sovereignty efforts across Europe, India, and Southeast Asia.
Anthropic IPO Timeline at Risk — Fable 5 Shutdown Hits Revenue, Investor Confidence, and Regulatory Story
The Fable 5 shutdown arrives at the worst possible moment for Anthropic's IPO planning. The company filed its confidential S-1 on June 1, targeting an October 2026 listing at a $965 billion valuation. The shutdown creates three immediate problems: lost enterprise revenue from disabled enterprise accounts; damaged "reliable partner" reputation with enterprise customers who lost access mid-deployment; and an unpredictable regulatory narrative that public market investors were not pricing in. Analysts note the US government applied stricter standards to Anthropic than to OpenAI's similarly capable GPT-5.5 — which Anthropic itself cited as containing the same jailbreak. Anthropic said it is working to restore access "as soon as possible" and believes the shutdown stems from a "misunderstanding." The company is simultaneously suing the Pentagon over a supply chain blacklist, negotiating with the Commerce Department over Fable 5, and trying to finalize IPO paperwork — all at the same time.
Apple Wraps WWDC — macOS Golden Gate Drops Intel Support Entirely, Xcode 27 Goes Agentic
Apple closed WWDC 2026 on June 12 with the full release of macOS Golden Gate — the first macOS version to drop Intel Mac support entirely. Intel machines will receive security patches only going forward. Every 2026 Apple OS (iOS 27, iPadOS 27, watchOS 27, tvOS 27, visionOS 27, macOS Golden Gate) ships with AI Intelligence features woven throughout via a redesigned Siri and the new View Annotations API. Xcode 27 is now 30% smaller, Apple Silicon-only, and ships agentic coding capabilities: it can interact with the iOS Simulator, run and fix failing tests, localize apps, and resolve crash logs autonomously. Apple Foundation Models is getting a free tier for developers with under 2 million App Store downloads and will go open-source later this summer. The unified API lets Swift apps call Claude, Gemini, and other third-party models through a single interface.
Google Gemini 3.5 Pro Nearing General Availability — 2M Token Context, Deep Think Reasoning
Google confirmed Gemini 3.5 Pro is nearing general availability following Sundar Pichai's Google I/O promise to "give us until next month." The model features a 2-million-token context window, a "Deep Think" reasoning mode, and native multimodal support across text, images, audio, and video. Gemini 3.5 Pro consolidates workloads previously routed to the Ultra tier — including the hardest reasoning tasks and longest-context analysis — while consumer subscribers on the $250/month Ultra plan get Deep Think access. Gemini 3.5 Flash, which shipped at Google I/O on May 19, already scores 76.2% on Terminal-Bench 2.1 and 83.6% on MCP Atlas — higher than Gemini 3.1 Pro on both. If Pro extends Flash's performance improvement proportionally, the competitive implications for Claude Opus and GPT-5.5 are significant.
Ramp Raises $750M at $44B Valuation to Build AI Finance Infrastructure
Corporate fintech Ramp closed a $750 million Series F led by ICONIQ, GIC, and Ontario Teachers' Pension Plan, lifting its valuation to $44 billion — up from $32 billion seven months ago. Revenue run rate: $1.5 billion with positive free cash flow. Ramp serves 70,000+ enterprise accounts including Visa, Uber, and Shopify. The company is doubling down on AI agents for expense management, procurement, and accounting, and launched a new corporate credit card built specifically for AI agents alongside a token-spend management product that helps enterprises track and control AI usage costs across providers. Ramp is also uniquely positioned as an AI market data source — its Ramp AI Index has become one of the most-cited trackers of enterprise AI adoption, recently confirming Anthropic overtook OpenAI in US business spend.
Claude Code Ships Nested Sub-Agents With Five-Level Deep Spawning
Claude Code version 2.1.170, released the week of June 10, introduced nested sub-agents — a capability allowing agents to spawn their own sub-agents up to five levels deep in a single session. This enables specialized parallel execution: a parent agent can simultaneously delegate to sub-agents handling tests, documentation, code review, and dependency analysis. The update also ships smarter model and region routing, a new plugin search system, and improved Chrome, VS Code, and terminal integration. The release reinforces Claude Code's position as the fastest-growing product in Anthropic's history — currently responsible for an estimated 4% of all public GitHub commits worldwide, double the share from one month prior.