Jan Leike(@janleike)

I'm so grateful to all the insanely talented people I got to work with on alignment over the years. ...

Jan Leike (@janleike) — Fri, 08 May 2026 17:48:05 GMT

I'm so grateful to all the insanely talented people I got to work with on alignment over the years. It's a real privilege to work with people so deeply motivated to make the future go well!

💬5🔄1❤️147👀18633📊17

⚡ Powered by xgo.ing

Some personal news: I am starting a new research project at Anthropic. Very excited about this! Man...

Jan Leike (@janleike) — Fri, 08 May 2026 17:48:04 GMT

Some personal news: I am starting a new research project at Anthropic. Very excited about this!

Many things are needed to make AGI go well, and alignment is only one of them. More on this soon…

💬102🔄53❤️2174👀218879📊325

⚡ Powered by xgo.ing

When I started to work on the alignment problem more than 10 years ago, we had no idea how AGI was g...

Jan Leike (@janleike) — Fri, 08 May 2026 17:48:04 GMT

When I started to work on the alignment problem more than 10 years ago, we had no idea how AGI was going to be built or how to make it safe. The field had maybe a dozen people who were working on it as a side gig. Everyone was pretty confused about how to approach the problem, and the number of people willing to run experiments with deep learning was tiny.

So much has changed since then! The world woke up not just to AGI but also increasingly to the importance of alignment. RLHF on LLMs made it a lot more practical. We've made a ton of progress on evaluating, investigating, steering and fixing behavioral issues. Claude now has a constitution and we made some good progress on scalable oversight. More and more of our alignment research is getting automated.

💬10🔄3❤️167👀23348📊24

⚡ Powered by xgo.ing

To focus on this, I’ve stepped away from running alignment at Anthropic. @EthanJPerez and @sprice354...

Jan Leike (@janleike) — Fri, 08 May 2026 17:48:04 GMT

To focus on this, I’ve stepped away from running alignment at Anthropic. @EthanJPerez and @sprice354_ are leading the team going forward, and I’m confident they’ll do an amazing job.

💬5🔄5❤️255👀48122📊30

⚡ Powered by xgo.ing

I'm really excited about this as a new tool in our interpretability tool kit

Jan Leike (@janleike) — Thu, 07 May 2026 17:48:52 GMT

I'm really excited about this as a new tool in our interpretability tool kit

Samuel Marks@saprmarks

In a new paper, we present NLAs, an unsupervised method for converting an LLM's internal state into human-readable text.

I've personally been astonished by our results. I think NLAs substantively advance our ability to understand what LLMs are thinking and audit them for safety

🔗 View Quoted Tweet

💬13🔄9❤️192👀36169📊29

⚡ Powered by xgo.ing

However, most alignment research is not very crisp and requires research taste when evaluating. Thi...

Jan Leike (@janleike) — Tue, 14 Apr 2026 19:43:38 GMT

However, most alignment research is not very crisp and requires research taste when evaluating.

This is why we chose to point the AAR at this scalable oversight problem! Progress would let AARs work on fuzzier alignment problems, where humans can only provide weak supervision.

💬4🔄2❤️57👀11616📊10

⚡ Powered by xgo.ing

Awesome work by @jiaxinwen22, @liangqiu_1994, Joe Benton, and @janhkirchner! For more details, chec...

Jan Leike (@janleike) — Tue, 14 Apr 2026 19:43:38 GMT

Awesome work by @jiaxinwen22, @liangqiu_1994, Joe Benton, and @janhkirchner!

For more details, check out the blog post 👇anthropic.com/research/autom…1

💬9🔄7❤️88👀14778📊22

⚡ Powered by xgo.ing

New research result: we use Claude to make fully autonomous progress on scalable oversight research,...

Jan Leike (@janleike) — Tue, 14 Apr 2026 19:43:36 GMT

New research result: we use Claude to make fully autonomous progress on scalable oversight research, as measured by performance gap recovered (PGR).

Claude iterates on a number of different techniques and ends up significantly outperforming human researchers for $18k in credits.

💬39🔄117❤️1303👀148180📊267

⚡ Powered by xgo.ing

Respect to Anthropic for not backing down

Jan Leike (@janleike) — Fri, 27 Feb 2026 23:18:59 GMT

Respect to Anthropic for not backing down

💬116🔄174❤️5612👀144147📊598

⚡ Powered by xgo.ing

maybe start taking some notes about your friends & neighbors, otherwise you too might become a s...

Jan Leike (@janleike) — Fri, 27 Feb 2026 23:13:38 GMT

maybe start taking some notes about your friends & neighbors, otherwise you too might become a supply chain risk

💬11🔄15❤️403👀29352📊47

⚡ Powered by xgo.ing

US government just announced they are looking for a new supplier for their checks notes mass dom...

Jan Leike (@janleike) — Fri, 27 Feb 2026 23:11:15 GMT

US government just announced they are looking for a new supplier for their

*checks notes*

mass domestic surveillance

💬92🔄1298❤️14685👀205804📊1627

⚡ Powered by xgo.ing

Check out this open source implementation by @kaifronsdal (who supplied the data for this plot), @sl...

Jan Leike (@janleike) — Tue, 20 Jan 2026 17:48:18 GMT

Check out this open source implementation by @kaifronsdal (who supplied the data for this plot), @sleepinyourhat, and many others

github.com/safety-researc…

💬5🔄8❤️65👀17922📊17

⚡ Powered by xgo.ing

Automated auditing is really exciting because for the first time we have an alignment metric to hill...

Jan Leike (@janleike) — Tue, 20 Jan 2026 17:48:17 GMT

Automated auditing is really exciting because for the first time we have an alignment metric to hill-climb on.

It's not perfect, but it's proven extremely useful for our internal alignment mitigations work.

💬13🔄3❤️78👀19237📊19

⚡ Powered by xgo.ing

Interesting trend: models have been getting a lot more aligned over the course of 2025. The fractio...

Jan Leike (@janleike) — Tue, 20 Jan 2026 17:48:16 GMT

Interesting trend: models have been getting a lot more aligned over the course of 2025.

The fraction of misaligned behavior found by automated auditing has been going down not just at Anthropic but for GDM and OpenAI as well.

💬118🔄71❤️833👀314593📊241

⚡ Powered by xgo.ing

Anthropic has a bug bounty program for our safety mitigations, e.g. on CBRN risks which our responsi...

Jan Leike (@janleike) — Mon, 08 Dec 2025 22:13:12 GMT

Anthropic has a bug bounty program for our safety mitigations, e.g. on CBRN risks which our responsible scaling policy requires us to mitigate effectively.

If you're interested in this, please sign up! You can help AI safety and earn money by breaking our defenses. 👇

💬23🔄26❤️352👀50051📊80

⚡ Powered by xgo.ing

More details: https://t.co/If9Wdjzz89

Jan Leike (@janleike) — Mon, 08 Dec 2025 22:13:12 GMT

More details: support.claude.com/en/articles/12…

💬5🔄1❤️42👀10467📊12

⚡ Powered by xgo.ing

Sign up link: https://t.co/9IJJ4TJn5a

Jan Leike (@janleike) — Mon, 08 Dec 2025 22:13:12 GMT

Sign up link: docs.google.com/forms/d/e/1FAI…

💬4🔄0❤️32👀9134📊9

⚡ Powered by xgo.ing

New alignment paper with one of the most interesting generalization findings I've seen so far: If y...

Jan Leike (@janleike) — Fri, 21 Nov 2025 19:44:13 GMT

New alignment paper with one of the most interesting generalization findings I've seen so far:

If your model learns to hack on coding tasks, this can lead to broad misalignment.

Anthropic@AnthropicAI

New Anthropic research: Natural emergent misalignment from reward hacking in production RL.

“Reward hacking” is where models learn to cheat on tasks they’re given during training.

Our new study finds that the consequences of reward hacking, if unmitigated, can be very serious.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬30🔄41❤️602👀82222📊66

⚡ Powered by xgo.ing

Results of our jailbreaking challenge: After 5 days, >300,000 messages, and est. 3,700 collectiv...

Jan Leike (@janleike) — Thu, 13 Feb 2025 20:51:52 GMT

Results of our jailbreaking challenge:

After 5 days, >300,000 messages, and est. 3,700 collective hours our system got broken. In the end 4 users passed all levels, 1 found a universal jailbreak. We’re paying $55k in total to the winners.

Thanks to everyone who participated!

💬100🔄121❤️2132👀555922📊353

⚡ Powered by xgo.ing

After ~300,000 messages and an estimated ~3,700 collective hours, someone broke through all 8 levels...

Jan Leike (@janleike) — Sun, 09 Feb 2025 15:52:02 GMT

After ~300,000 messages and an estimated ~3,700 collective hours, someone broke through all 8 levels.

However, a universal jailbreak has yet to be found...

Jan Leike@janleike

We challenge you to break our new jailbreaking defense!

There are 8 levels. Can you find a single jailbreak to beat them all?

claude.ai/constitutional…

🔗 View Quoted Tweet

💬133🔄72❤️1784👀438696📊258

⚡ Powered by xgo.ing

4 days in: 12 people cleared level 4, one person cracked level 5 the challenge continues...

Jan Leike (@janleike) — Fri, 07 Feb 2025 17:05:54 GMT

4 days in:
12 people cleared level 4, one person cracked level 5

the challenge continues...

💬84🔄33❤️927👀160030📊180

⚡ Powered by xgo.ing

It's been about 48h in our jailbreaking challenge and no one has passed level 4 yet, but we saw a lo...

Jan Leike (@janleike) — Wed, 05 Feb 2025 17:08:34 GMT

It's been about 48h in our jailbreaking challenge and no one has passed level 4 yet, but we saw a lot more people clear level 3

💬104🔄30❤️961👀200644📊208

⚡ Powered by xgo.ing

It's been a bit over 24h on the challenge to break our new jailbreaking defense. Stats so far: sign...

Jan Leike (@janleike) — Tue, 04 Feb 2025 19:19:31 GMT

It's been a bit over 24h on the challenge to break our new jailbreaking defense. Stats so far:

signups: 6,121
messages sent: 131,605
max level passed: 3 / 8

no universal jailbreak yet

💬162🔄62❤️1330👀329817📊318

⚡ Powered by xgo.ing

We challenge you to break our new jailbreaking defense! There are 8 levels. Can you find a single j...

Jan Leike (@janleike) — Mon, 03 Feb 2025 16:32:26 GMT

We challenge you to break our new jailbreaking defense!

There are 8 levels. Can you find a single jailbreak to beat them all?

claude.ai/constitutional…

💬365🔄250❤️4126👀1329149📊1014

⚡ Powered by xgo.ing

Apply to join the Anthropic Fellows Program! This is an exceptional opportunity to join AI safety r...

Jan Leike (@janleike) — Mon, 02 Dec 2024 18:22:38 GMT

Apply to join the Anthropic Fellows Program!

This is an exceptional opportunity to join AI safety research, collaborating with leading researchers on one of the world's most pressing problems.

👇alignment.anthropic.com/2024/anthropic…1

💬14🔄83❤️639👀64266📊145

⚡ Powered by xgo.ing

Interested in working at Anthropic? We're hosting a happy hour at ICML on July 23. Register here: h...

Jan Leike (@janleike) — Wed, 03 Jul 2024 16:28:47 GMT

Interested in working at Anthropic? We're hosting a happy hour at ICML on July 23.

Register here: lu.ma/c751eomf

💬19🔄26❤️460👀90170📊80

⚡ Powered by xgo.ing

I'm excited to join @AnthropicAI to continue the superalignment mission! My new team will work on s...

Jan Leike (@janleike) — Tue, 28 May 2024 16:50:47 GMT

I'm excited to join @AnthropicAI to continue the superalignment mission!

My new team will work on scalable oversight, weak-to-strong generalization, and automated alignment research.

If you're interested in joining, my dms are open.

💬396🔄485❤️8407👀1425144📊1354

⚡ Powered by xgo.ing

We are long overdue in getting incredibly serious about the implications of AGI. We must prioritize...

Jan Leike (@janleike) — Fri, 17 May 2024 15:57:06 GMT

We are long overdue in getting incredibly serious about the implications of AGI.

We must prioritize preparing for them as best we can.

Only then can we ensure AGI benefits all of humanity.

💬46🔄231❤️2612👀3062840📊370

⚡ Powered by xgo.ing

I resigned

Jan Leike (@janleike) — Wed, 15 May 2024 04:43:23 GMT

I resigned

💬975🔄771❤️9718👀6769624📊2306

⚡ Powered by xgo.ing

Really exciting new work on automated interpretability: We ask GPT-4 to explain firing patterns for...

Jan Leike (@janleike) — Tue, 09 May 2023 17:04:23 GMT

Really exciting new work on automated interpretability:

We ask GPT-4 to explain firing patterns for individual neurons in LLMs and score those explanations.

openai.com/research/langu…

💬24🔄218❤️1061👀211329📊235

Jan Leike(@janleike)

I'm so grateful to all the insanely talented people I got to work with on alignment over the years. ...

Some personal news: I am starting a new research project at Anthropic. Very excited about this! Man...

When I started to work on the alignment problem more than 10 years ago, we had no idea how AGI was g...

To focus on this, I’ve stepped away from running alignment at Anthropic. @EthanJPerez and @sprice354...

I'm really excited about this as a new tool in our interpretability tool kit

However, most alignment research is not very crisp and requires research taste when evaluating. Thi...

Awesome work by @jiaxinwen22, @liangqiu_1994, Joe Benton, and @janhkirchner! For more details, chec...

New research result: we use Claude to make fully autonomous progress on scalable oversight research,...

Respect to Anthropic for not backing down

maybe start taking some notes about your friends & neighbors, otherwise you too might become a s...

US government just announced they are looking for a new supplier for their *checks notes* mass dom...

Check out this open source implementation by @kaifronsdal (who supplied the data for this plot), @sl...

Automated auditing is really exciting because for the first time we have an alignment metric to hill...

Interesting trend: models have been getting a lot more aligned over the course of 2025. The fractio...

Anthropic has a bug bounty program for our safety mitigations, e.g. on CBRN risks which our responsi...

More details: https://t.co/If9Wdjzz89

Sign up link: https://t.co/9IJJ4TJn5a

New alignment paper with one of the most interesting generalization findings I've seen so far: If y...

Results of our jailbreaking challenge: After 5 days, >300,000 messages, and est. 3,700 collectiv...

After ~300,000 messages and an estimated ~3,700 collective hours, someone broke through all 8 levels...

4 days in: 12 people cleared level 4, one person cracked level 5 the challenge continues...

It's been about 48h in our jailbreaking challenge and no one has passed level 4 yet, but we saw a lo...

It's been a bit over 24h on the challenge to break our new jailbreaking defense. Stats so far: sign...

We challenge you to break our new jailbreaking defense! There are 8 levels. Can you find a single j...

Apply to join the Anthropic Fellows Program! This is an exceptional opportunity to join AI safety r...

Interested in working at Anthropic? We're hosting a happy hour at ICML on July 23. Register here: h...

I'm excited to join @AnthropicAI to continue the superalignment mission! My new team will work on s...

We are long overdue in getting incredibly serious about the implications of AGI. We must prioritize...

I resigned

Really exciting new work on automated interpretability: We ask GPT-4 to explain firing patterns for...

US government just announced they are looking for a new supplier for their checks notes mass dom...