PromptHacker Weekly | Issue #129

Anthropic's Best Models Went Dark. Copilot Cowork Took Over the Work.

Fable 5 went dark three days after launch, Copilot Cowork now finishes the task instead of drafting it, and Grok learned to make video. Here is what is worth your Wednesday.

Welcome

The strangest AI story of the year happened on Friday. Three days after Anthropic put Fable 5 in your plan, the US government ordered it pulled, and Anthropic disabled both Fable 5 and Mythos 5 for everyone.

Opus 4.8 still handles focused document work. Microsoft Copilot Cowork now completes multi-step assignments, and Grok Imagine turns a product still into a short video. Below, each section gives you one practical test for work, health, or family.

01 / Quick Hits

OpenAI retired GPT-5.2 from ChatGPT. GPT-5.5 is now the default, so re-pin any saved model preference and test your two most-used prompts before the next deadline. Use one real workflow, not a sample. Read more

Claude usage moved to a separate credit pool. Check your monthly ceiling and update any scripts still pinned to retired Claude 4 model IDs. Run one automated task before relying on it. Read more

Gemini in Chrome is coming to Android. The mobile rollout can summarize pages, answer questions about the screen, generate images, and auto-browse for eligible paid users. Test it on one long article first. Read more

Grok tightened its shared usage limits. Text, image, voice, code, and video now draw from one cumulative ceiling, so heavy media generation can throttle later work. Read more

Anthropic launched Claude Corps. The $150 million program starts with 100 fellows helping nonprofits build practical Claude workflows, with plans to scale toward 1,000 fellows. Read more

02 / Top AI Updates

1. Anthropic's two most powerful models went offline

Three days after launch, Anthropic received a federal export-control directive and disabled Claude Fable 5 and Mythos 5 for every user. The restriction covered foreign-national access, and Anthropic could not separate affected users in real time. Opus 4.8 and Sonnet 4.6 remained available. Anthropic said it was working with regulators, but no restoration date was available.

Why it matters: A critical workflow can disappear with one policy decision. Treat model availability as concentration risk, especially for contracts, research, and long-document analysis. A tested fallback is basic workflow design, not a technical luxury.

Action items:

• Move active Fable 5 work to Opus 4.8.
• Split very long documents into focused batches.
• Keep one qualified fallback model for critical work.

Read the full analysis

2. Copilot Cowork now finishes the whole assignment

Microsoft made Copilot Cowork generally available worldwide. It runs long, multi-tool tasks in the cloud and returns a finished artifact instead of a draft. Cowork stays inside the Microsoft 365 security boundary, requires a Copilot license, and uses metered credits after an administrator enables it. Tasks continue when the laptop closes, using Work IQ context and approved files.

Why it matters: The useful unit of work changes from helping with one step to completing a defined assignment. That can remove hours of analyst work when the review standard is clear. The gain only counts when the finished artifact survives human review without heavy cleanup.