/Tech3h ago

Google DeepMind's Andreas Kirsch argues AI researchers can no longer ignore the downstream consequences of advancing model capabilities

Story Overview

Andreas Kirsch's recent tweet captures a personal shift at DeepMind, where the once straightforward pursuit of stronger models now feels tangled with questions about who controls their eventual use, especially after governance changes and deals that stayed opaque to staff.

110211.2K

#271

Original post

Andreas Kirsch 🇺🇦@BlackHC#271inTech

While it lasted, I really really enjoyed not worrying about the consequences of making AI more capable 🥺

it was such a nice life where I could just focus on research. I wish and hope I can go back to that soon and trust again that I can rely on others to take care of these worries

but when there is no independent oversight and actual AGI governance and AI principles get modified to later sign opaque deals with reportedly no apparent enforceable guardrails and without telling employees about it or the details, then it's really difficult to choose to remain ignorant, not worry and "carry on" 🥲

Miles Brundage@Miles_Brundage

It must be so fun to spend all day making AI more capable + not worrying about the consequences

IMO you should not do that — at the very least you should push for good policies, inform the public, speed up good uses etc. a bit if not switch to safety — but man that must be fun

2:44 AM · Jun 19, 2026 · 1K Views

Policy Risk

Pentagon contract sparks researcher pushback

Kirsch publicly labeled the April 2026 classified AI work with the Defense Department as a breach of earlier standards, noting it caught employees off guard and left no clear guardrails in place.

Open Question

Pure capability era looks harder to reclaim

The thread leaves open whether stronger independent oversight will appear soon enough for researchers to stop tracking downstream effects themselves.

Sentiment

Users dismiss AI safety concerns as a weaponized framing against open source, arguing that advancing capability alone is the real priority.

Pos

0.0%

Neg

100.0%

1 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

Gene Kogan@genekogan

@Miles_Brundage It doesn’t seem like many people work on capability and safety at the same time. I kind of think most people understand they are not the difference maker on whether the consequences turn out good or bad. If they genuinely think it’s the latter, they stop working on capability.

Miles Brundage@Miles_Brundage

It must be so fun to spend all day making AI more capable + not worrying about the consequences

IMO you should not do that — at the very least you should push for good policies, inform the public, speed up good uses etc. a bit if not switch to safety — but man that must be fun

2h18700

Nathan Quantum@AI_WarriorNQ

@BlackHC making AI more capable IS the good use. this whole 'worry about consequences' framing just gets weaponized to kill open source and lock access behind corporate gates

2h1