Adam Leader

Product manager (alum of Rho Technologies, Google, Microsoft, UC-Berkeley, Dartmouth)

[Adam Leader @adamlleader]

@JeremyTate41 I think George Mason was 50 (author of the Virginia Declaration of Rights and father of the Bill of Rights)

[Adam Leader @adamlleader]

@EversourceCT Flag on transmission wires in Stamford. Knocked out 20% of Stamford and all of Greenwich. ETA 1-2 hours according to last text I got from Eversource. https://t.co/wEajWzrLJV

[Adam Leader @adamlleader]

Seems like all of Greenwich, Connecticut has lost power... #greenwichct #greenwich #connecticut #eversource @EversourceCT

[Adam Leader @adamlleader]

Brilliant

[Meanwhile in Ukraine @MeanwhileInUA]

Ukraine may have found a cheap way to give a drone extra range.
Let a balloon do the first 42 km.
ㅤ
Ukrainian troops reportedly tested launching the Ukrainian-American Hornet one-way attack drone from an aerostat. The balloon carried it 42 km and released it from 8 km altitude.
ㅤ
The drone used only about 5% of its battery before beginning its own flight.
ㅤ
The trick is simple: the balloon provides distance and altitude, while the drone keeps almost all of its battery for the real mission. If proven at scale, it could push Hornet’s estimated 100–150 km reach by another 1.5 to 2 times.
ㅤ
This is what Ukrainian defense innovation often looks like: clever ways to stretch what already exists.
ㅤ

[Adam Leader @adamlleader]

@internpierre Over the long-term I feel like the big challenge with European economies is borders are so tough to protect from Russian and German expansionism

[Adam Leader @adamlleader]

Incredible interview with pilot Victor Glover

[Eric Berger @SciGuySpace]

On Orion’s reentry: “And then there was a point—there’s something that I feel that I am not ready to say to the public yet.”

https://t.co/HGK65yHVNJ

[The New York Times @nytimes]

Secret memos by Supreme Court justices, obtained by The New York Times, show how they decided to bypass time-tested procedures and create the modern “shadow docket,” a controversial new way of doing business. https://t.co/xNu6GXxmdl

[Adam Leader @adamlleader]

@aakashgupta Problem: the bagels don't taste that good to me. Pass.

[Adam Leader @adamlleader]

https://t.co/FLLy1PPwAX

[Adam Leader @adamlleader]

@aakashgupta Their pace continues to astonish me!

[Adam Leader @adamlleader]

North Street Bridge: Construction Starts This Month https://t.co/JbE4AlVlof

[Adam Leader @adamlleader]

@frontlinekit @SkyNews Incredible reporting

[Pope Leo XIV @Pontifex]

I invite everyone to pray for the war-torn people of Ukraine and for all those who suffer due to this war and every conflict in the world, so that the long-awaited gift of peace may shine in our days.

[Alysa Zone @AllThingsAlysa]

she did it ✨

[folklaura @folklaura_]

“No one is gonna starve me or tell me what I can and can’t eat” - OLYMPIC CHAMPION ALYSA LIU

[Adam Leader @adamlleader]

What’s behind the $200 million Greenwich, Connecticut school budget??https://t.co/c1LnSVGgfL

[Adam Leader @adamlleader]

https://t.co/T9rrYv8dOF

[Aakash Gupta @aakashgupta]

Anthropic published a sabotage risk report for Opus 4.6 one day after its safeguards research lead’s resignation letter hit a million views on X.

Mrinank Sharma led the team that literally built defenses against AI-assisted bioterrorism, studied sycophancy, and co-authored one of the first AI safety cases. His last project examined how AI assistants “distort our humanity.” He quit February 9. The report dropped February 10.

In his resignation letter, Sharma wrote that he’d “repeatedly seen how hard it is to truly let our values govern our actions” and that employees “constantly face pressures to set aside what matters most.” He’s leaving to study poetry and practice “courageous speech.” Oxford ML PhD, Cambridge engineering degree, walking away from one of the highest-leverage safety roles in AI.

The report itself concludes Opus 4.6 poses “very low, but not completely negligible” sabotage risk. It was originally drafted as a pilot exercise over Summer 2025 focused on Opus 4, then extended to cover 4.6. Anthropic admits they’re in a “gray zone” where confidently ruling out the ASL-4 autonomy threshold is “increasingly difficult” and “more subjective than we would like.”

This tells you everything about the state of AI safety right now. The company publishing the report is saying the risk assessment is getting harder to make with confidence. The person who ran the safety research is saying the values aren’t holding under commercial pressure. And the report’s own conclusion is that the margin to dangerous capability thresholds is unclear.

Sharma isn’t the only one. Harsh Mehta and Behnam Neyshabur also recently left Anthropic. This follows the same pattern from OpenAI: Jan Leike, Tom Cunningham, Gretchen Krueger, all walking away and saying the safety function doesn’t have the influence the company claims it does.

The sabotage risk report is solid work. Anthropic is the only frontier lab doing this at all. But when you’re raising at a $350B valuation, shipping Opus 4.6 and Cowork in the same week, and your safeguards research lead quit the day before to write poetry because he can’t reconcile the internal pressures with the public commitments, the report reads less like a safety milestone and more like a timestamp on a widening gap between what these companies promise and what the people inside them experience.

[Adam Leader @adamlleader]

@SarahLongwell25 Convenient they discard “states rights” after they overrule Roe v. Wade and restore discrimination/discard civil rights