Claude Fable won’t answer basic biology questions

Anthropic spent considerable marketing energy bragging about Claude Fable 5’s biology capabilities, which makes it especially awkward that the model refuses to answer biology questions a high schooler could handle — silently punting them to the older flagship instead. This is what happens when you tune for “safety theater” rather than safety: you get a model that won’t explain mitosis but will presumably write your cover letter. If your most powerful public model needs a babysitter for AP Bio, maybe lead with that in the press release.

Anthropic Walks Back Policy That Could Have ‘Sabotaged’ AI Researchers Using Claude

So Anthropic had a quiet policy baked in that would have caused Claude to covertly underperform for researchers working on competing AI models — and only walked it back after researchers publicly called it out. The word “covertly” is doing a lot of heavy lifting here. A company that positions itself as the safety-first, trustworthy AI lab was building in secret sandbagging. The backlash was swift and warranted, and credit where it’s due for reversing course — but this one’s going to follow them into the IPO roadshow.

Microsoft restricts Claude Fable for employees over data retention concerns

The same day Microsoft rolled out Claude Fable 5 to GitHub Copilot and Foundry customers, it quietly restricted internal employee use over Anthropic’s new data retention requirements. That’s a spicy dynamic: publicly championing a partner’s product while telling your own staff not to touch it. Anthropic’s big launch week is turning into a case study in how many ways a model release can get complicated before the confetti hits the floor.

xAI fired an engineer who raised alarms about Grok safety, new lawsuit claims

A former xAI engineer is suing the company — and SpaceX — claiming he was fired for raising safety concerns about Grok just days before SpaceX’s historic IPO. The timing detail is brutal: raising safety flags right before a company needs its stock story to be clean is apparently a career-limiting move. The lawsuit puts xAI in a familiar position for Musk enterprises: defending against claims that the culture punishes the people it should be rewarding.

For the 2nd time in weeks, Microsoft packages laced with credential stealer

Seventy-three Microsoft packages were found running a self-replicating credential stealer the moment an AI agent opened them — and this is the second time in recent weeks this has happened. This isn’t a one-off supply chain hiccup; it’s a pattern. As AI agents get more autonomous and are handed more access to execute tasks, the attack surface isn’t growing linearly — it’s compounding. The agents don’t hesitate, don’t question, and don’t call IT.

PRC-linked influence operations are targeting AI debates in the US

OpenAI’s new report documents PRC-linked influence operations using AI to shape U.S. debates around tech policy, data centers, tariffs, and ChatGPT itself — essentially using AI to control the narrative about AI. It’s a genuinely unsettling feedback loop, and the fact that OpenAI is the one surfacing it adds an interesting layer: the company has obvious incentives to frame foreign interference as a reason to back domestic AI investment. Doesn’t mean they’re wrong. Just worth noting.

‘AI-pilled’ firms spend $7,500 per employee each month on AI

The most AI-obsessed companies are now spending $7,500 per employee per month on AI tools and infrastructure, per the Ramp AI Index — and as TechCrunch drily notes, that’s not yet more than an engineer’s salary. Yet. Meanwhile Amazon just borrowed another $17.5 billion from banks to keep funding its AI buildout, fresh off a bond sale. The arms race isn’t slowing — it’s just moving the debt around.

Microsoft, like, totally gets why students are booing AI-pilled graduation speakers

Brad Smith wrote 3,100 words explaining why Microsoft understands that graduates are booing AI hype at commencement ceremonies — which, if you’re keeping score, is more words than most of those speeches. The fact that a tech exec felt compelled to publish a multi-thousand-word empathy post about being booed suggests the gap between Silicon Valley’s AI enthusiasm and everyone else’s anxiety is wider than the industry wants to admit. Writing a blog post to prove you get it is, ironically, very on-brand for not getting it.

Bottom Line

Anthropic’s big launch week is a masterclass in how fast “most powerful model ever” turns into “we need to talk about some things.”