Key Takeaways
- Anthropic's Fable 5 returned after 19 days — but free usage caps change July 7; check your team's billing
- AI security research flagged a 29-year-old Squid proxy vulnerability humans had missed since 1997
- HP is standardizing company-wide on OpenAI after pilots showed one engineer doing team-scale output
- Even Amazon and Anthropic are fighting over AI pricing — assume your costs will move, not stay fixed
- Build workflows around what the system does, not which vendor provides it this month
Week of June 29 – July 5, 2026
Three weeks ago we covered a top AI model pulled by export controls days after launch. This week it came back — with different pricing. Same week: an AI found a security hole that survived 29 years of human review, HP bet its entire operation on a single AI vendor, and two of the largest AI companies on earth started fighting over a bill.
This is Last Week in AI — the signal, not the noise. Four stories operators need to understand, and what each one means for how you run your business.
Story 1: Fable 5 is back — read the fine print before July 7
The US Commerce Department lifted export controls on June 30. Fable 5 returned to users worldwide on July 1.
The catch: Fable 5 is only included at no extra cost for up to half of a subscriber's weekly usage through July 7. After that, using it requires paying for usage credits on top of the subscription you're already paying for.
What it means for operators: The model is back, but "back" doesn't mean "free like before." If your team built anything around this tool while it was included, check the billing change before you keep using it the same way. The pricing that got you in is not the pricing you're on now.
Story 2: AI found a bug that survived 29 years of human review
A vulnerability in Squid — one of the most widely deployed proxy servers in the world — was discovered this week by an AI model running through Anthropic's security research program. The bug exposed user credentials to any attacker on the same network. It had passed every human code review, security audit, and penetration test since it was written in 1997.
The same research program has flagged over 23,000 vulnerabilities across 1,000 open-source projects in a single month, with a 90.6% confirmation rate on independent verification.
What it means for operators: This isn't a chatbot writing emails. This is AI doing work entire human teams could not do economically at this depth. If your business runs any software — website, booking system, payment processor — the vendors behind those tools now have access to this level of review. Whether they're using it is worth asking.
Story 3: HP is standardizing on one AI vendor company-wide
HP announced an expanded company-wide OpenAI partnership this week, citing results from earlier pilots:
- One engineer reportedly processed 122 pull requests across 43 projects in a matter of weeks using AI tools
- A security team fixed bugs in a day that HP estimates would have taken up to a month manually
HP is moving from pilot programs to making one AI vendor its core infrastructure company-wide.
What it means for operators: A company the size of HP went all in after seeing results like that from a pilot. You don't need HP's budget to get a version of that outcome — you need someone to identify which single workflow in your business would benefit most and build around it the same way.
Story 4: Amazon and Anthropic are fighting over AI pricing
Anthropic moved Amazon to a new token-based billing structure under a renegotiated contract — a change reported to sharply increase what Amazon pays. Amazon holds an $8 billion stake in Anthropic now worth an estimated $74 billion, and remains its primary cloud provider. Amazon also disputes that its costs are actually rising.
The tension sits on top of an already strange relationship: Amazon's CEO was involved in the export controls that pulled Anthropic's models offline in the first place.
What it means for operators: If the company that owns nearly a tenth of Anthropic is fighting with them over pricing, the rates you're paying as a small customer are not stable either. Nobody's contract in this industry right now is a fixed number. Build with the assumption that your AI costs will change — because the people at the top are proving it constantly.
This week's principle
A model came back with different pricing than it left with. An AI outperformed 29 years of human security review. A hardware giant bet everything on one vendor. Two AI companies are fighting over a bill.
None of that is stable ground — and none of it needs to be, if you build around what the system does for you rather than which company is providing it this month.
What to do this week
If anyone on your team has been using AI tools heavily, check usage limits and billing before July 7. Limits changed with almost no warning — twice — in under a month. That's not a reason to stop using the tools. It's a reason to build workflows so a single vendor's pricing change doesn't take down your whole process.
Not sure where your stack is fragile? Start with a free ops assessment — we'll map where manual work and vendor dependency are costing you time.