🏠 Fable is back... kind of

GA! It's Brett.

Apologies for the delays. Had a tough week with my former exec coach passing away. She was an incredible person. You can read about her here.

On the lighter side of things, we released an awesome native LinkedIn integration in Micro this week. You can check it out for free by signing up below.

Anthropic's Fable is back. On June 30, the U.S. lifted the export controls that had blocked Anthropic from making its Mythos and Fable models broadly available abroad, and Anthropic said it would begin restoring access on July 1.

Here is what makes the story truly interesting. Anthropic did not ship an unconstrained model. Fable 5 was designed from the start with hard safety limits — when a user asks it to help with biology, cybersecurity, distillation, or anything in those danger zones, Fable doesn't just refuse and dead-end you. Classifiers catch the request and hand it off to Opus 4.8, Anthropic's more cautious model, which answers instead. The architecture had constraint built in from day one.

The trigger for the shutdown was a specific jailbreak discovered by Amazon researchers — a technique that prompted Fable 5 to identify software vulnerabilities and, in one case, produce exploit code. After the shutdown, Anthropic trained an additional classifier that blocks the technique in more than 99% of cases.

Then the U.S. government added both Mythos and Fable to its export-control list on June 12, three days after launch, saying they were too dangerous to share abroad. The irony is sharp: the U.S. restricted a model that Anthropic had already architected to refuse the most dangerous requests. Fable does not make bioweapons or write exploits. And the jailbreak that triggered the ban? It worked on GPT-5.5 too.

That tells you something crucial about what the controls are actually for. This is not about technical capability. It is about access and leverage. The U.S. did not restrict Fable because Fable is dangerous. It restricted Fable because Fable is good, and good frontier models are now treated as strategic assets. Anthropic built safety into the model. The government built controls around it anyway. That is the new pattern.

The AI tool you tried last quarter waited for a prompt, hallucinated a number, then asked if you'd like a summary.

Viktor opened a PR at 2am, rebased it against main, ran your test suite, and posted a note in #eng: "Two flaky tests in payments service, both pre-existing. Recommended merging after fixing them." Then drafted the customer reply for the support ticket the bug created.

That's 619K autonomous actions per day across 20,000+ teams. Not chat replies. Real work shipped to GitHub, Stripe, Linear, Notion, and 3,000+ other tools, from inside Slack and Microsoft Teams.

You don't supervise him any more than you supervise a senior engineer.

SOC 2 certified. Your data never trains models.

"It's what you probably originally thought AI was going to be when you first heard of it in sci-fi movies." Tyler, CEO.

Get Started for Free

From prompts to production: Startup multi-agent workflows

One of the more useful videos from this past week is a June 25 YouTube session on how startups are moving from single prompts to real multi-agent systems. The core idea is simple: stop treating AI like a smarter chat box and start designing small chains of responsibility, where one agent researches, another structures, and a third executes. YouTube: Startup multi-agent workflows

That framing feels even more relevant after this week's Fable reversal. Better models matter, but the bigger unlock is still operational. The teams that win will be the ones that can build durable systems on top of shifting model access and changing economics. YouTube: Startup multi-agent workflows TechCrunch: restrictions dropped

Watch the full video

5 things on my mind this week:

Venice AI raised a $65 million Series A. The privacy-first AI platform runs models locally on your device — no cloud, no data collection. The round signals investors are still willing to pay up for opinionated product positioning in a market dominated by giant labs. Privacy as a feature, not a footnote. — TechCrunch

Meta is facing $1.4 trillion in penalties in the teen mental health case — nearly its entire valuation. 4.1M views. @nypost

A single white rose. Black background.

Last week's poll: Left was real — a sunset airplane photo. Only 30.4% of you got it right. 39.1% voted Right, 21.7% voted Neither, 8.7% voted Both. Real life just looks like AI now.

Micro is a CRM that doesn’t feel like homework and an email client that knows what matters. Simply put, it’s an all-in-one productivity tool that organizes itself so you don’t have to.

LHV is a community-driven, early-stage venture fund backing ambitious founders at the earliest stages.

If you’re building tools that help founders, investors and other professionals who want to be 100x more productive with 100x less work, and you’re interested in advertising with us, send an email over to [email protected] with the subject “Homescreen Ads”