AI Safety Papers

Sharing the latest in AI safety research.

[AI Safety Papers @safe_paper]

Log analysis is necessary for credible evaluation of AI agents

Peter Kirgis, Sayash Kapoor (@sayashk), Stephan Rabanser (@steverab), Nitya Nadgir, Cozmin Ududec (@CUdudec), Magda Dubois (@DubMagda), JJ Allaire (@fly_upside_down), @MariusHobbhahn, Jacob Steinhardt (@JacobSteinhar2), Arvind Narayanan (@random_walker)

[AI Safety Papers @safe_paper]

https://t.co/D89BFpBrM1

[AI Safety Papers @safe_paper]

Automated alignment is harder than you think

Aleksandr Bowkis (@aleksandrbowkis), Marie Davidsen Buhl (@MarieBassBuhl), @jacob_pfau, Geoffrey Irving (@geoffreyirving)

@AISecurityInst

[AI Safety Papers @safe_paper]

https://t.co/UkVVWUbMGs

[AI Safety Papers @safe_paper]

Emergent Misalignment is Easy, Narrow Misalignment is Hard

Anna Soligo (@anna_soligo), Edward Turner, Senthooran Rajamanoharan (@sen_r), Neel Nanda (@NeelNanda5)

[AI Safety Papers @safe_paper]

https://t.co/ltdiMPdvqv

[AI Safety Papers @safe_paper]

Distributional AGI Safety

Nenad Tomašev (@weballergy), Matija Franklin (@FranklinMatija), Julian Jacobs (@JulianDJacobs), Sébastien Krier (@sebkrier), Simon Osindero (@sindero)

@GoogleDeepMind

[AI Safety Papers @safe_paper]

https://t.co/1A6ATUEeq2

[AI Safety Papers @safe_paper]

Legal Alignment for Safe and Ethical AI

Noam Kolt, Nicholas Caputo, Jack Boeglin, Cullen O'Keefe, @RishiBommasani, @StephenLCasper, Mariano-Florentino Cuéllar, @profnoahfeldman, @IasonGabriel, Gillian K. Hadfield (@ghadfield), Lewis Hammond (@lrhammond), Peter Henderson (@PeterHndrsn), Atoosa Kasirzadeh (@Dr_Atoosa), @sethlazar, @AnkaReuel, @kevinlwei, Jonathan Zittrain (@zittrain)

[AI Safety Papers @safe_paper]

https://t.co/YgEoXxFtES

[AI Safety Papers @safe_paper]

Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs

Jan Betley (@BetleyJan), @JorioCocola, Dylan Feng (@dylanfeng_), James Chua (@jameschua_sg), Andy Arditi (@andyarditi), Anna Sztyber-Betley (@anna_sztyber), Owain Evans (@OwainEvans_UK)

[AI Safety Papers @safe_paper]

https://t.co/DrpMltC5Us

[AI Safety Papers @safe_paper]

Natural Emergent Misalignment from Reward Hacking in Production RL

Monte MacDiarmid, Benjamin Wright (@RightBenguin), @JonathanUesato, @JoeJBenton, Jon Kutasov, Sara Price (@sprice354_), Naia Bouscal, Sam Bowman (@sleepinyourhat), @TrentonBricken, Alex Cloud, Carson Denison, Johannes Gasteiger (@gasteigerjo), @RyanPGreenblatt, @janleike, @Jack_W_Lindsey, Vlad Mikulik, @EthanJPerez, @alexrodriguesca, Drake Thomas (@MaskedTorah), @albertwebson, Daniel Ziegler (@d_m_ziegler), Evan Hubinger (@EvanHub)

@AnthropicAI @redwood_ai

[AI Safety Papers @safe_paper]

https://t.co/CoJTVDx6cg

[AI Safety Papers @safe_paper]

A dataset of rated conceptual arguments

Caspar Oesterheld (@C_Oesterheld), Emery Cooper, Linh Chi Nguyen, Alexander Kastner, @EthanJPerez

[AI Safety Papers @safe_paper]

https://t.co/oBLCQTcSxt

[AI Safety Papers @safe_paper]

Quantifying Elicitation of Latent Capabilities in Language Models

Elizabeth Donoway, @HaileyJoren, Arushi Somani, Henry Sleight (@sleight_henry), @_julianmichael_ , Michael R DeWeese, John Schulman (@johnschulman2), @EthanJPerez, @FabienDRoger, @janleike

@AnthropicAI

[AI Safety Papers @safe_paper]

https://t.co/XdOQrTpyjj

[AI Safety Papers @safe_paper]

Remote Labor Index: Measuring AI Automation of Remote Work
Mantas Mazeika (@MantasMazeika96), Alice Gatti, Cristina Menghini (@CriMenghini), Udari Madhushani Sehwag, Shivam Singhal (@ShivamSinghal56), Yury Orlovskiy (@yvorlovskiy), [...], Summer Yue (@summeryue0), @alexandr_wang, Bing Liu (@vbingliu), Ernesto Hernandez (@eghmontoya), @hendrycks

@cais @scale_AI