A Commerce Dept. stop letter gave Anthropic 90 minutes; Fable 5 / Mythos 5 went dark across all channels with no migration window.

Until last week, Fable 5 was live and in active use. It was designated under export controls with zero transition time and no deprecation notice.

Regulatory risk now sits beside infrastructure failure. Teams running AI-dependent pipelines need a backup-model plan in place this week.

Filed in D.C. federal court: Max 5x ($100/mo) and Max 20x ($200/mo) plans don't reach the advertised usage multipliers.

Max plan usage has always been token-weighted, not raw message count — a detail that wasn't front and center when users signed up for higher tiers.

If the case proceeds, Anthropic may need to revise how it describes Max tier limits. Heavy users who've felt shortchanged should watch this.

Anthropic dropped its planned per-credit charge for the Claude Agent SDK; apps and third-party integrations continue on existing subscription limits.

Two weeks ago, a credit-based SDK billing model was announced, raising concerns among third-party developers about higher integration costs ahead.

Good news for builders — SDK stays free within your current plan. Anthropic cited price competition with OpenAI as the reason for the reversal.

76 cybersecurity experts sent an open letter to the White House urging rollback of Fable 5 / Mythos 5 export restrictions.

Since the export ban last month, industry pushback had been quiet. This open letter marks the first organized expert response against the restrictions.

The outcome will determine whether the Anthropic API stays fully accessible outside the U.S. Risk for non-U.S. teams using Fable-class models continues.

SearchLeak (CVE-2026-42824) chains prompt injection → HTML race condition → SSRF to exfiltrate data via a single link click. Microsoft patched server-side.

Until recently, Copilot prompt-injection risks were treated as theoretical. This CVE gives them a concrete attack chain and published proof-of-concept.

Patch is live, but confirm your tenant is updated. Expect similar chained attack patterns to appear more often as Copilot adoption grows.

A critical-severity prompt injection let attackers retrieve 2FA codes from Copilot-accessible emails. Microsoft patched the underlying guardrail bypass.

Last month, Copilot email processing was considered safe. This flaw showed LLMs cannot reliably separate user instructions from malicious content in emails.

Patched now, but the root problem — LLMs acting on behalf of users while processing untrusted content — isn't going away anytime soon.

Copilot Cowork — Microsoft's answer to Claude Cowork — went GA for M365 Copilot users on June 16, shifting from flat fee to usage-based billing.

The March trial ran on fixed pricing, but agent workloads scale unpredictably. Usage-based billing lets teams pay for what they actually consume.

Easier to justify in procurement — no large upfront commitment. Enterprises evaluating AI agents now have a per-use option alongside subscriptions.

SpaceX officially announced the ~$60B acquisition of Cursor (Anysphere) on June 16, just after its historic IPO; Q3 2026 close targeted per SEC filings.

A preliminary deal was hinted at in April, but no formal announcement came. SpaceX's IPO created the capital base and strategic urgency to close it.

SpaceX is now a serious player in enterprise AI coding via xAI. Cursor users should watch for roadmap shifts as corporate priorities change.

Sensor Tower's State of AI Report 2026 shows ChatGPT's market share dipped below 50% for the first time, with users moving to Gemini, Claude, and Grok.

Six months ago, ChatGPT held a dominant majority in the AI assistant market. Rivals' usability has improved enough to accelerate meaningful switching.

The 'everyone uses ChatGPT' default is fading in business. Tool selection now requires deliberate comparison — which works in favor of challengers.

DOJ sided with xAI in the Memphis gas-turbine permit case, arguing that cutting power would damage national security and disrupt military AI operations.

Last month, an environmental lawsuit sought to shut down xAI's unpermitted Memphis turbines. Active government backing of private AI infrastructure at this scale is unprecedented.

If AI infrastructure is officially a national-security asset, maintenance beats regulation. This framing could spread to other countries' AI policy debates.

Digital Agency released AI Procurement Guidelines v2.0: a CAIO at each ministry plus a detailed 54-item checklist. Governance from July 1; full rollout Sept 1.

Previous guidelines focused on text-based AI and didn't cover voice or image inputs clearly. There was no standard CAIO role defined across ministries.

Vendors selling AI to Japanese government agencies must meet the new checklist by September 1. AI governance frameworks kick in July 1, ahead of the rest.

ElevenLabs partnered with Tooin (Japan's veteran translation firm) for end-to-end enterprise dubbing: transcription → translation → TTS → delivery in one contract.

Until now, Japanese enterprises wanting AI dubbing at scale had to self-assemble TTS, translation QA, and delivery. There was no single-vendor option.

Companies localizing video content to or from Japanese now have a turn-key B2B option with a professional translation firm's quality guarantee behind it.

Anthropic's updated privacy policy (effective July 8) adds age and identity verification for Free/Pro/Max users; Persona handles ID, Yoti handles age checks.

Until now, Claude accounts only required an email address. From July 8, selected users may be asked for a photo ID and a live selfie to continue.

Users wary of ID verification will face a higher bar to continue. Impact on API users and enterprise developers appears limited at this stage.