Legion, a US-based litigation-technology company whose development team includes Canadian employees working remotely from Canada, filed suit in Washington federal court on June 23 challenging the Commerce Department directive that forced Anthropic to disable Fable 5 and Mythos 5 globally on June 12 (Bloomberg, The Next Web). The Bureau of Industry and Security order required Anthropic to block access by any foreign national; because the API layer has no mechanism to verify nationality in real time, Anthropic shut both models off for all users worldwide (AI Weekly). Anthropic had publicly launched Fable 5 on June 9 - it runs at $10 per million input tokens and $50 per million output tokens on the API - three days before the order halted access (TechCrunch, Finout). Legion’s complaint, described as the first customer legal challenge to an AI model export ban, argues the competitive harm of losing frontier-model access during a period of rapid advancement is “immediate, irreparable, and existential,” and asks the court to vacate the directive (Gizmodo).
Mistral AI released OCR 4 on June 23, a document-extraction model that emits bounding boxes, typed-block labels, and per-word confidence scores as structured output alongside the recognized text (Mistral). The system supports 170 languages in a single containerized deployment designed as a self-hosted ingestion layer for RAG and agentic search pipelines, with API pricing at $4 per 1,000 pages for standard jobs and $2 per 1,000 pages for batch workloads (MarkTechPost). Independent annotators preferred OCR 4 over every competing document-AI system in blind tests at a 72% average win rate, and the model recorded the top score on OlmOCRBench at 85.20 (TechTimes).
Also on June 23, Anthropic launched Claude Tag, a persistent Slack agent available via @Claude in any channel for Claude Enterprise and Team subscribers (Fortune). The integration allows teams to assign multi-step tasks directly in Slack threads; Claude Tag retains context across channels it is granted access to and proactively surfaces updates without being prompted (The Next Web). Anthropic noted that an internal build of the tool now generates 65% of its own product team’s code, and cited Ramp’s May AI index showing Anthropic had pulled ahead of OpenAI in US business adoption for the first time, at 34.4% versus 32.3% of surveyed firms (TechTimes).