https://www.matsprogram.org https://www.matsprogram.org/about https://www.matsprogram.org/contact https://www.matsprogram.org/faq https://www.matsprogram.org/research https://www.matsprogram.org/privacy https://www.matsprogram.org/apply https://www.matsprogram.org/search https://www.matsprogram.org/alumni https://www.matsprogram.org/mentors https://www.matsprogram.org/team https://www.matsprogram.org/transparency https://www.matsprogram.org/careers https://www.matsprogram.org/faq/getting-into-mats https://www.matsprogram.org/donate https://www.matsprogram.org/contact/success https://www.matsprogram.org/substack https://www.matsprogram.org/careers/compute-administrator-berkeley https://www.matsprogram.org/careers/head-of-people https://www.matsprogram.org/careers/recruiting-lead https://www.matsprogram.org/careers/research-manager-berkeley-london https://www.matsprogram.org/careers/senior-research-manager-berkeley-london https://www.matsprogram.org/alumni/aaron-sher https://www.matsprogram.org/alumni/aashiq-muhamed https://www.matsprogram.org/alumni/adam-karvonen https://www.matsprogram.org/alumni/aengus-lynch https://www.matsprogram.org/alumni/alan-cooney https://www.matsprogram.org/alumni/andis-draguns https://www.matsprogram.org/alumni/anish-mudide https://www.matsprogram.org/alumni/ann-kathrin-dombrowski https://www.matsprogram.org/alumni/aram-ebtekar https://www.matsprogram.org/alumni/aryan-bhatt https://www.matsprogram.org/alumni/asa-cooper-stickland https://www.matsprogram.org/alumni/asa-cooper-strckland https://www.matsprogram.org/alumni/bilal-chugthai https://www.matsprogram.org/alumni/callum-mcdougall https://www.matsprogram.org/alumni/can-rager https://www.matsprogram.org/alumni/casey-barkan https://www.matsprogram.org/alumni/claire-short https://www.matsprogram.org/alumni/cody-rushing https://www.matsprogram.org/alumni/constantin-weisser https://www.matsprogram.org/alumni/damiano-fornasiere https://www.matsprogram.org/alumni/dan-valentine https://www.matsprogram.org/alumni/dane-sherburn https://www.matsprogram.org/alumni/daniel-tan https://www.matsprogram.org/alumni/david-abecassis https://www.matsprogram.org/alumni/david-matolcsi https://www.matsprogram.org/alumni/david-udell https://www.matsprogram.org/alumni/dillon-bowen https://www.matsprogram.org/alumni/egg-syntax https://www.matsprogram.org/alumni/even-wybitul https://www.matsprogram.org/alumni/govind-pampali https://www.matsprogram.org/alumni/guang-yuan-hao https://www.matsprogram.org/alumni/hantao-lou https://www.matsprogram.org/alumni/hoagy-cunningham https://www.matsprogram.org/alumni/jai-dhyani https://www.matsprogram.org/alumni/jakub-krys https://www.matsprogram.org/alumni/jay-bailey https://www.matsprogram.org/alumni/jesse-hoogland https://www.matsprogram.org/alumni/jessica-rumbelow https://www.matsprogram.org/alumni/jiaxin-wen https://www.matsprogram.org/alumni/johannes-treutlein https://www.matsprogram.org/alumni/john-hughes https://www.matsprogram.org/alumni/jonah-weinbaum https://www.matsprogram.org/alumni/joseph-bloom https://www.matsprogram.org/alumni/julius-steen https://www.matsprogram.org/alumni/kai-fronsdal https://www.matsprogram.org/alumni/kay-kozaronek https://www.matsprogram.org/alumni/lilly-stelling https://www.matsprogram.org/alumni/lisa-thiergart https://www.matsprogram.org/alumni/lucie-philippon https://www.matsprogram.org/alumni/marcus-williams https://www.matsprogram.org/alumni/marius-hobbhahn https://www.matsprogram.org/alumni/matt-macdermott https://www.matsprogram.org/alumni/matthew-farr https://www.matsprogram.org/alumni/megan-kinniment https://www.matsprogram.org/alumni/michael-pearce https://www.matsprogram.org/alumni/nina-panickssery https://www.matsprogram.org/alumni/oliver-sourbut https://www.matsprogram.org/alumni/oly-sourbut https://www.matsprogram.org/alumni/paul-lorxus-rapoport https://www.matsprogram.org/alumni/paul-reichers https://www.matsprogram.org/alumni/peter-barnett https://www.matsprogram.org/alumni/quentin-feuillade-montixi https://www.matsprogram.org/alumni/rajashree-agrawal https://www.matsprogram.org/alumni/rauno-arike https://www.matsprogram.org/alumni/rebecca-hawkins-and-steven-veld https://www.matsprogram.org/alumni/robert-krzyzanowski https://www.matsprogram.org/alumni/rupal-jain https://www.matsprogram.org/alumni/rupali-bhati https://www.matsprogram.org/alumni/sam-marks https://www.matsprogram.org/alumni/satvik-golechha https://www.matsprogram.org/alumni/sunishchal-dev https://www.matsprogram.org/alumni/teun-van-der-weij https://www.matsprogram.org/alumni/thomas-kwa https://www.matsprogram.org/alumni/thomas-larsen https://www.matsprogram.org/alumni/wes-gurnee https://www.matsprogram.org/program/autumn-2026 https://www.matsprogram.org/program/summer-2022 https://www.matsprogram.org/program/summer-2023 https://www.matsprogram.org/program/summer-2024 https://www.matsprogram.org/program/summer-2025 https://www.matsprogram.org/program/summer-2026 https://www.matsprogram.org/program/winter-2023 https://www.matsprogram.org/program/winter-2024 https://www.matsprogram.org/program/winter-2025 https://www.matsprogram.org/program/winter-2026 https://www.matsprogram.org/research/a-benchmark-for-scalable-oversight-protocols https://www.matsprogram.org/research/a-causal-model-of-theory-of-mind-in-ai-agents https://www.matsprogram.org/research/activation-oracles-training-and-evaluating-llms-as-general-purpose-activation-explainers https://www.matsprogram.org/research/adaptive-attacks-on-trusted-monitors-subvert-ai-control-protocols https://www.matsprogram.org/research/adaptive-deployment-of-untrusted-llms-reduces-distributed-threats https://www.matsprogram.org/research/adaptive-sparse-allocation-with-mutual-choice-feature-choice-sparse-autoencoders https://www.matsprogram.org/research/adversarial-circuit-evaluation https://www.matsprogram.org/research/agentic-misalignment-how-llms-could-be-insider-threats https://www.matsprogram.org/research/agentic-property-based-testing-finding-bugs-across-the-python-ecosystem https://www.matsprogram.org/research/ai-researchers-views-on-automating-ai-r-d-and-intelligence-explosions https://www.matsprogram.org/research/ai-sandbagging-language-models-can-strategically-underperform-on-evaluations https://www.matsprogram.org/research/among-us-a-sandbox-for-measuring-and-detecting-agentic-deception https://www.matsprogram.org/research/analyzing-probabilistic-methods-for-evaluating-agent-capabilities https://www.matsprogram.org/research/applying-sparse-autoencoders-to-unlearn-knowledge-in-language-models https://www.matsprogram.org/research/are-sparse-autoencoders-useful-a-case-study-in-sparse-probing https://www.matsprogram.org/research/audit-cards-contextualizing-ai-evaluations https://www.matsprogram.org/research/auditing-language-models-for-hidden-objectives https://www.matsprogram.org/research/automatically-finding-reward-model-biases https://www.matsprogram.org/research/badllama-cheaply-removing-safety-fine-tuning-from-llama-2-chat-13b https://www.matsprogram.org/research/base-models-know-how-to-reason-thinking-models-learn-when https://www.matsprogram.org/research/batchtopk-sparse-autoencoders https://www.matsprogram.org/research/believe-it-or-not-how-deeply-do-llms-believe-implanted-facts https://www.matsprogram.org/research/best-of-n-jailbreaking https://www.matsprogram.org/research/bias-augmented-consistency-training-reduces-biased-reasoning-in-chain-of-thought https://www.matsprogram.org/research/biases-in-the-blind-spot-detecting-what-llms-fail-to-mention https://www.matsprogram.org/research/bilinear-mlps-enable-weight-based-mechanistic-interpretability https://www.matsprogram.org/research/bloom-an-open-source-tool-for-automated-behavioral-evaluations https://www.matsprogram.org/research/can-aha-moments-be-fake-identifying-true-and-decorative-thinking-steps-in-chain-of-thought https://www.matsprogram.org/research/can-llms-follow-simple-rules https://www.matsprogram.org/research/can-reasoning-models-obfuscate-reasoning-stress-testing-chain-of-thought-monitorability https://www.matsprogram.org/research/catastrophic-goodhart-regularizing-rlhf-with-kl-divergence-does-not-mitigate-heavy-tailed-reward-misspecification https://www.matsprogram.org/research/censored-llms-as-a-natural-testbed-for-secret-knowledge-elicitation https://www.matsprogram.org/research/chain-of-thought-reasoning-in-the-wild-is-not-always-faithful https://www.matsprogram.org/research/claudini-autoresearch-discovers-state-of-the-art-adversarial-attack-algorithms-for-llms https://www.matsprogram.org/research/compact-proofs-of-model-performance-via-mechanistic-interpretability https://www.matsprogram.org/research/constitutional-black-box-monitoring-for-scheming-in-llm-agents https://www.matsprogram.org/research/constrained-belief-updates-explain-geometric-structures-in-transformer-representations https://www.matsprogram.org/research/control-tax-the-price-of-keeping-ai-in-check https://www.matsprogram.org/research/convergent-linear-representations-of-emergent-misalignment https://www.matsprogram.org/research/copy-suppression-comprehensively-understanding-an-attention-head https://www.matsprogram.org/research/ctrl-z-controlling-ai-agents-via-resampling https://www.matsprogram.org/research/debating-with-more-persuasive-llms-leads-to-more-truthful-answers https://www.matsprogram.org/research/detecting-adversarial-fine-tuning-with-auditing-agents https://www.matsprogram.org/research/diff-mining-logit-differences-reveal-finetuning-objectives https://www.matsprogram.org/research/difr-inference-verification-despite-nondeterminism https://www.matsprogram.org/research/distillation-robustifies-unlearning https://www.matsprogram.org/research/distributed-and-decentralised-training-technical-governance-challenges-in-a-shifting-ai-landscape https://www.matsprogram.org/research/do-i-know-this-entity-knowledge-awareness-and-hallucinations-in-language-models https://www.matsprogram.org/research/do-the-rewards-justify-the-means-measuring-trade-offs-between-rewards-and-ethical-behavior-in-the-machiavelli-benchmark https://www.matsprogram.org/research/eliciting-harmful-capabilities-by-fine-tuning-on-safeguarded-outputs https://www.matsprogram.org/research/eliciting-secret-knowledge-from-language-models https://www.matsprogram.org/research/emergent-misalignment-is-easy-narrow-misalignment-is-hard https://www.matsprogram.org/research/emergent-misalignment-narrow-finetuning-can-produce-broadly-misaligned-llms https://www.matsprogram.org/research/estimating-the-empowerment-of-language-model-agents https://www.matsprogram.org/research/evaluating-sparse-autoencoders-on-targeted-concept-erasure-tasks https://www.matsprogram.org/research/expert-selections-in-moe-models-reveal-almost-as-much-as-text https://www.matsprogram.org/research/exploration-hacking-can-llms-learn-to-resist-rl-training https://www.matsprogram.org/research/exploration-hacking-llms-can-learn-to-resist-rl-training https://www.matsprogram.org/research/explorations-of-self-repair-in-language-models https://www.matsprogram.org/research/failures-to-find-transferable-image-jailbreaks-between-vision-language-models https://www.matsprogram.org/research/feature-hedging-correlated-features-break-narrow-sparse-autoencoders https://www.matsprogram.org/research/finding-neurons-in-a-haystack-case-studies-with-sparse-probing https://www.matsprogram.org/research/forecasting-frontier-language-model-agent-capabilities https://www.matsprogram.org/research/frontier-models-can-take-actions-at-low-probabilities https://www.matsprogram.org/research/future-events-as-backdoor-triggers-investigating-temporal-vulnerabilities-in-llms https://www.matsprogram.org/research/goodhart-s-law-in-reinforcement-learning https://www.matsprogram.org/research/gradient-routing-masking-gradients-to-localize-computation-in-neural-networks https://www.matsprogram.org/research/haiku-to-opus-in-just-10-bits-llms-unlock-massive-compression-gains https://www.matsprogram.org/research/higher-order-belief-in-incomplete-information-maids https://www.matsprogram.org/research/how-does-information-access-affect-llm-monitors-ability-to-detect-sabotage https://www.matsprogram.org/research/how-to-catch-an-ai-liar-lie-detection-in-black-box-llms-by-asking-unrelated-questions https://www.matsprogram.org/research/identifying-functionally-important-features-with-end-to-end-sparse-dictionary-learning https://www.matsprogram.org/research/identifying-sparsely-active-circuits-through-local-loss-landscape-decomposition https://www.matsprogram.org/research/improving-steering-vectors-by-targeting-sparse-autoencoder-features https://www.matsprogram.org/research/incentivizing-honest-performative-predictions-with-proper-scoring-rules https://www.matsprogram.org/research/inoculation-prompting-eliciting-traits-from-llms-during-training-can-suppress-them-at-test-time https://www.matsprogram.org/research/interpbench-semi-synthetic-transformers-for-evaluating-mechanistic-interpretability-techniques https://www.matsprogram.org/research/interpretability-as-compression-reconsidering-sae-explanations-of-neural-activations-with-mdl-saes https://www.matsprogram.org/research/interpreting-language-model-parameters https://www.matsprogram.org/research/is-this-the-subspace-you-are-looking-for-an-interpretability-illusion-for-subspace-activation-patching https://www.matsprogram.org/research/language-models-learn-to-mislead-humans-via-rlhf https://www.matsprogram.org/research/large-language-models-can-learn-and-generalize-steganographic-chain-of-thought-under-process-supervision https://www.matsprogram.org/research/large-language-models-often-know-when-they-are-being-evaluated https://www.matsprogram.org/research/large-scale-online-deanonymization-with-llms https://www.matsprogram.org/research/latent-adversarial-training-improves-robustness-to-persistent-harmful-behaviors-in-llms https://www.matsprogram.org/research/learning-multi-level-features-with-matryoshka-sparse-autoencoders https://www.matsprogram.org/research/limitations-of-agents-simulated-by-predictive-models https://www.matsprogram.org/research/linear-representations-of-sentiment-in-large-language-models https://www.matsprogram.org/research/litmusvalues-will-ai-tell-lies-to-save-sick-children-litmus-testing-ai-values-prioritization-with-airiskdilemmas https://www.matsprogram.org/research/llm-evaluators-recognize-and-favor-their-own-generations https://www.matsprogram.org/research/lora-fine-tuning-efficiently-undoes-safety-training-in-llama-2-chat-70b https://www.matsprogram.org/research/mapping-industry-practices-to-the-eu-ai-act-s-gpai-code-of-practice-safety-and-security-measures https://www.matsprogram.org/research/me-myself-and-ai-the-situational-awareness-dataset-sad-for-llms https://www.matsprogram.org/research/measuring-chain-of-thought-monitorability-through-faithfulness-and-verbosity https://www.matsprogram.org/research/measuring-control-intervention-awareness-across-frontier-llms https://www.matsprogram.org/research/measuring-progress-in-dictionary-learning-for-language-model-interpretability-with-board-game-models https://www.matsprogram.org/research/misr-measuring-instrumental-self-reasoning-in-frontier-models https://www.matsprogram.org/research/mitigating-reward-hacking-with-rl-training-interventions https://www.matsprogram.org/research/model-organisms-for-emergent-misalignment https://www.matsprogram.org/research/model-tampering-attacks-enable-more-rigorous-evaluations-of-llm-capabilities https://www.matsprogram.org/research/modular-training-of-neural-networks-aids-interpretability https://www.matsprogram.org/research/more-capable-less-cooperative-when-llms-fail-at-zero-cost-collaboration https://www.matsprogram.org/research/morebench-evaluating-procedural-and-pluralistic-moral-reasoning-in-language-models-more-than-outcomes https://www.matsprogram.org/research/narrow-finetuning-leaves-clearly-readable-traces-in-activation-differences https://www.matsprogram.org/research/neural-chameleons-language-models-can-learn-to-hide-their-thoughts-from-unseen-activation-monitors https://www.matsprogram.org/research/on-defining-neural-averaging https://www.matsprogram.org/research/on-targeted-manipulation-and-deception-when-optimizing-llms-for-user-feedback https://www.matsprogram.org/research/open-character-training-shaping-the-persona-of-ai-assistants-through-constitutional-ai https://www.matsprogram.org/research/optimizing-ai-agent-attacks-with-synthetic-data https://www.matsprogram.org/research/output-supervision-can-obfuscate-the-chain-of-thought https://www.matsprogram.org/research/overcoming-sparsity-artifacts-in-crosscoders-to-interpret-chat-tuning https://www.matsprogram.org/research/petri-an-open-source-auditing-tool-to-accelerate-ai-safety-research https://www.matsprogram.org/research/planning-in-a-recurrent-neural-network-that-plays-sokoban https://www.matsprogram.org/research/probabilistic-modeling-of-latent-agentic-substructures-in-deep-neural-networks https://www.matsprogram.org/research/probing-by-analogy-decomposing-probes-into-activations-for-better-interpretability-and-inter-model-generalization https://www.matsprogram.org/research/prompt-optimization-makes-misalignment-legible https://www.matsprogram.org/research/public-perspectives-on-ai-governance-a-survey-of-working-adults-in-california-illinois-and-new-york https://www.matsprogram.org/research/quantifying-stability-of-non-power-seeking-in-artificial-agents https://www.matsprogram.org/research/rapid-response-mitigating-llm-jailbreaks-with-a-few-examples https://www.matsprogram.org/research/re-bench-evaluating-frontier-ai-r-d-capabilities-of-language-model-agents-against-human-experts https://www.matsprogram.org/research/real-time-detection-of-hallucinated-entities-in-long-form-generation https://www.matsprogram.org/research/reasoning-finetuning-repurposes-latent-representations-in-base-models https://www.matsprogram.org/research/reasoning-models-struggle-to-control-their-chains-of-thought https://www.matsprogram.org/research/reasoning-under-pressure-how-do-training-incentives-influence-chain-of-thought-monitorability https://www.matsprogram.org/research/rec1JwTkhydMN5WE7 https://www.matsprogram.org/research/recFhdiMsgY0JL2JU https://www.matsprogram.org/research/recMzkohGHIh6PXEz https://www.matsprogram.org/research/reccQQWZti5eBtQKV https://www.matsprogram.org/research/recd2EkuzuD97qNAD https://www.matsprogram.org/research/recdMJOyz2zmhhlQt https://www.matsprogram.org/research/recfNoXMnEKY0AGIp https://www.matsprogram.org/research/recgHlR4BY1n4Xmrm https://www.matsprogram.org/research/recj3n9EyIrLn4WqI https://www.matsprogram.org/research/recja5R4wRTraVmb8 https://www.matsprogram.org/research/recjyXchMfcmRdogM https://www.matsprogram.org/research/recl3jesjpfz1TNPp https://www.matsprogram.org/research/recl7OoHazTbgCHoL https://www.matsprogram.org/research/reclAAaZK3Xrt1qBX https://www.matsprogram.org/research/reclEhSOBqZfVqP5K https://www.matsprogram.org/research/reclWdAVUuKLCeJcr https://www.matsprogram.org/research/recontextualization-mitigates-specification-gaming-without-modifying-the-specification https://www.matsprogram.org/research/recqktwhyKTrfjLAM https://www.matsprogram.org/research/rectiMaN6dbKi6YGO https://www.matsprogram.org/research/recus2SdRez7fNKKZ https://www.matsprogram.org/research/recvGgl86ama10sxo https://www.matsprogram.org/research/recxapWTc3zIf1FCd https://www.matsprogram.org/research/recxiS2xpWReqCnqF https://www.matsprogram.org/research/refusal-in-language-models-is-mediated-by-a-single-direction https://www.matsprogram.org/research/removing-sandbagging-in-llms-by-training-with-weak-supervision https://www.matsprogram.org/research/representation-engineering-a-top-down-approach-to-ai-transparency https://www.matsprogram.org/research/resisting-rl-elicitation-of-biosecurity-capabilities-reasoning-models-exploration-hacking-on-wmdp https://www.matsprogram.org/research/rl-obfuscation-can-language-models-learn-to-evade-latent-space-monitors https://www.matsprogram.org/research/same-question-different-lies-cross-context-consistency-c3-for-black-box-sandbagging-detection https://www.matsprogram.org/research/secret-collusion-among-generative-ai-agents https://www.matsprogram.org/research/shifting-the-gradient-understanding-how-defensive-training-methods-protect-language-model-integrity https://www.matsprogram.org/research/simple-llm-baselines-are-competitive-for-model-diffing https://www.matsprogram.org/research/simple-mechanistic-explanations-for-out-of-context-reasoning https://www.matsprogram.org/research/sl5-standard-for-ai-security https://www.matsprogram.org/research/solidgoldmagikarp-plus-prompt-generation https://www.matsprogram.org/research/sparse-but-wrong-incorrect-l0-leads-to-incorrect-features-in-sparse-autoencoders https://www.matsprogram.org/research/steering-evaluation-aware-language-models-to-act-like-they-are-deployed https://www.matsprogram.org/research/steering-language-models-with-activation-engineering https://www.matsprogram.org/research/steering-language-models-with-weight-arithmetic https://www.matsprogram.org/research/steering-llama-2-via-contrastive-activation-addition https://www.matsprogram.org/research/steering-out-of-distribution-generalization-with-concept-ablation-fine-tuning https://www.matsprogram.org/research/stress-testing-alignment-audits-with-prompt-level-strategic-deception https://www.matsprogram.org/research/synthsaebench-evaluating-sparse-autoencoders-on-scalable-realistic-synthetic-data https://www.matsprogram.org/research/taken-out-of-context-on-measuring-situational-awareness-in-llms https://www.matsprogram.org/research/teaching-models-to-verbalize-reward-hacking-in-chain-of-thought-reasoning https://www.matsprogram.org/research/tell-don-t-show-declarative-facts-influence-how-llms-generalize https://www.matsprogram.org/research/the-2025-ai-agent-index https://www.matsprogram.org/research/the-assistant-axis-situating-and-stabilizing-the-default-persona-of-language-models https://www.matsprogram.org/research/the-attacker-moves-second-stronger-adaptive-attacks-bypass-defenses-against-llm-jailbreaks-and-prompt-injections https://www.matsprogram.org/research/the-partially-observable-off-switch-game https://www.matsprogram.org/research/the-wmdp-benchmark-measuring-and-reducing-malicious-use-with-unlearning https://www.matsprogram.org/research/thought-anchors-which-llm-reasoning-steps-matter https://www.matsprogram.org/research/thought-branches-interpreting-llm-reasoning-requires-resampling https://www.matsprogram.org/research/time-complexity-for-deterministic-string-machines https://www.matsprogram.org/research/too-late-to-recall-the-two-hop-problem-in-multimodal-knowledge-retrieval https://www.matsprogram.org/research/towards-a-unified-and-verified-understanding-of-group-operation-networks https://www.matsprogram.org/research/towards-eliciting-latent-knowledge-from-llms-with-mechanistic-interpretability https://www.matsprogram.org/research/towards-safeguarding-llm-fine-tuning-apis-against-cipher-attacks https://www.matsprogram.org/research/towards-understanding-sycophancy-in-language-models https://www.matsprogram.org/research/training-agents-to-self-report-misbehavior https://www.matsprogram.org/research/training-dynamics-of-contextual-n-grams-in-language-models https://www.matsprogram.org/research/training-large-language-models-on-narrow-tasks-can-lead-to-broad-misalignment https://www.matsprogram.org/research/transcoders-find-interpretable-llm-feature-circuits https://www.matsprogram.org/research/uncertainty-aware-policy-preserving-abstractions-with-abstention-for-one-shot-decisions https://www.matsprogram.org/research/understanding-and-controlling-a-maze-solving-policy-network https://www.matsprogram.org/research/understanding-reasoning-in-thinking-language-models-via-steering-vectors https://www.matsprogram.org/research/unelicitable-backdoors-in-language-models-via-cryptographic-transformer-circuits https://www.matsprogram.org/research/using-degeneracy-in-the-loss-landscape-for-mechanistic-interpretability https://www.matsprogram.org/research/verifying-llm-inference-to-detect-model-weight-exfiltration https://www.matsprogram.org/research/vista-dataset-do-vision-language-models-understand-sequential-tasks https://www.matsprogram.org/research/weird-generalization-and-inductive-backdoors-new-ways-to-corrupt-llms https://www.matsprogram.org/research/what-happens-when-superhuman-ais-compete-for-control https://www.matsprogram.org/research/what-should-frontier-ai-developers-disclose-about-internal-deployments https://www.matsprogram.org/research/when-just-read-the-chain-of-thought-fails-five-tasks-for-stress-testing-cot-monitors https://www.matsprogram.org/research/where-does-olmo-get-its-values https://www.matsprogram.org/research/where-is-the-mind-persona-vectors-and-llm-individuation https://www.matsprogram.org/research/will-ai-tell-lies-to-save-sick-children-litmus-testing-ai-values-prioritization-with-airiskdilemmas https://www.matsprogram.org/mentor/abel https://www.matsprogram.org/mentor/andriushchenko https://www.matsprogram.org/mentor/arnesen https://www.matsprogram.org/mentor/arx https://www.matsprogram.org/mentor/baker https://www.matsprogram.org/mentor/bengio https://www.matsprogram.org/mentor/benton https://www.matsprogram.org/mentor/binder https://www.matsprogram.org/mentor/bowman https://www.matsprogram.org/mentor/bricken https://www.matsprogram.org/mentor/butlin https://www.matsprogram.org/mentor/campbell https://www.matsprogram.org/mentor/campbell-17294 https://www.matsprogram.org/mentor/carlier https://www.matsprogram.org/mentor/carlini https://www.matsprogram.org/mentor/carroll https://www.matsprogram.org/mentor/casper https://www.matsprogram.org/mentor/chen https://www.matsprogram.org/mentor/choo https://www.matsprogram.org/mentor/cloud https://www.matsprogram.org/mentor/cohen https://www.matsprogram.org/mentor/conmy https://www.matsprogram.org/mentor/cooney https://www.matsprogram.org/mentor/cooper-stickland https://www.matsprogram.org/mentor/crook https://www.matsprogram.org/mentor/davies https://www.matsprogram.org/mentor/dean https://www.matsprogram.org/mentor/demski https://www.matsprogram.org/mentor/donoughe https://www.matsprogram.org/mentor/duan https://www.matsprogram.org/mentor/dvijotham https://www.matsprogram.org/mentor/egan https://www.matsprogram.org/mentor/emmons https://www.matsprogram.org/mentor/erwan https://www.matsprogram.org/mentor/etinger https://www.matsprogram.org/mentor/falet https://www.matsprogram.org/mentor/feng https://www.matsprogram.org/mentor/fields https://www.matsprogram.org/mentor/fish https://www.matsprogram.org/mentor/fornasiere https://www.matsprogram.org/mentor/gentzel https://www.matsprogram.org/mentor/giglemiani https://www.matsprogram.org/mentor/greenblatt https://www.matsprogram.org/mentor/grosse https://www.matsprogram.org/mentor/he https://www.matsprogram.org/mentor/heide https://www.matsprogram.org/mentor/hilton https://www.matsprogram.org/mentor/hobbhahn https://www.matsprogram.org/mentor/hoogland https://www.matsprogram.org/mentor/kang https://www.matsprogram.org/mentor/kaufman https://www.matsprogram.org/mentor/kinniment https://www.matsprogram.org/mentor/kirk https://www.matsprogram.org/mentor/kokotajlo https://www.matsprogram.org/mentor/korbak https://www.matsprogram.org/mentor/krakovna https://www.matsprogram.org/mentor/kristian-ronn https://www.matsprogram.org/mentor/kulp https://www.matsprogram.org/mentor/larsen https://www.matsprogram.org/mentor/lecomte https://www.matsprogram.org/mentor/lifland https://www.matsprogram.org/mentor/lindner https://www.matsprogram.org/mentor/lindsey https://www.matsprogram.org/mentor/majid https://www.matsprogram.org/mentor/mallen https://www.matsprogram.org/mentor/marks https://www.matsprogram.org/mentor/matthews https://www.matsprogram.org/mentor/mcaleer https://www.matsprogram.org/mentor/mccormick https://www.matsprogram.org/mentor/megastream https://www.matsprogram.org/mentor/merizian https://www.matsprogram.org/mentor/millman https://www.matsprogram.org/mentor/mir https://www.matsprogram.org/mentor/moorhouse https://www.matsprogram.org/mentor/mossing https://www.matsprogram.org/mentor/murfet https://www.matsprogram.org/mentor/nasr https://www.matsprogram.org/mentor/neel https://www.matsprogram.org/mentor/neyman https://www.matsprogram.org/mentor/ngo https://www.matsprogram.org/mentor/oehm https://www.matsprogram.org/mentor/parikh https://www.matsprogram.org/mentor/patel https://www.matsprogram.org/mentor/phuong https://www.matsprogram.org/mentor/price https://www.matsprogram.org/mentor/qi https://www.matsprogram.org/mentor/raushenbush https://www.matsprogram.org/mentor/richards https://www.matsprogram.org/mentor/richardson https://www.matsprogram.org/mentor/riechers https://www.matsprogram.org/mentor/robinson https://www.matsprogram.org/mentor/roger https://www.matsprogram.org/mentor/rondeau https://www.matsprogram.org/mentor/rushing https://www.matsprogram.org/mentor/sam https://www.matsprogram.org/mentor/schmitt https://www.matsprogram.org/mentor/schwettmann https://www.matsprogram.org/mentor/shai https://www.matsprogram.org/mentor/sharkey https://www.matsprogram.org/mentor/shlegeris https://www.matsprogram.org/mentor/siddiqui https://www.matsprogram.org/mentor/souly https://www.matsprogram.org/mentor/sourbut https://www.matsprogram.org/mentor/st-charles https://www.matsprogram.org/mentor/stastny https://www.matsprogram.org/mentor/steinhardt https://www.matsprogram.org/mentor/swett https://www.matsprogram.org/mentor/tour https://www.matsprogram.org/mentor/tracy https://www.matsprogram.org/mentor/trebacz https://www.matsprogram.org/mentor/trout https://www.matsprogram.org/mentor/turan https://www.matsprogram.org/mentor/turner https://www.matsprogram.org/mentor/ududec https://www.matsprogram.org/mentor/uribe https://www.matsprogram.org/mentor/varjavand https://www.matsprogram.org/mentor/veiga https://www.matsprogram.org/mentor/wang https://www.matsprogram.org/mentor/wang-fd161 https://www.matsprogram.org/mentor/warr https://www.matsprogram.org/mentor/westover https://www.matsprogram.org/mentor/wildeford https://www.matsprogram.org/mentor/winer https://www.matsprogram.org/mentor/winsor https://www.matsprogram.org/mentor/wolfe https://www.matsprogram.org/mentor/wu https://www.matsprogram.org/mentor/wu-38e19 https://www.matsprogram.org/mentor/zimmermann https://www.matsprogram.org/program-group/biosecurity https://www.matsprogram.org/program-group/empirical https://www.matsprogram.org/program-group/founding-and-field-building https://www.matsprogram.org/program-group/policy-and-governance https://www.matsprogram.org/program-group/strategy-and-forecasting https://www.matsprogram.org/program-group/systems-security https://www.matsprogram.org/program-group/theory https://www.matsprogram.org/stream/abel https://www.matsprogram.org/stream/abel-001b9 https://www.matsprogram.org/stream/aifp https://www.matsprogram.org/stream/aifp-bb5de https://www.matsprogram.org/stream/andriushchenko https://www.matsprogram.org/stream/anthropic https://www.matsprogram.org/stream/arc https://www.matsprogram.org/stream/arc-0c69d https://www.matsprogram.org/stream/baker https://www.matsprogram.org/stream/binder https://www.matsprogram.org/stream/binder-4b173 https://www.matsprogram.org/stream/butlin https://www.matsprogram.org/stream/campbell https://www.matsprogram.org/stream/carlier https://www.matsprogram.org/stream/carlier-4200f https://www.matsprogram.org/stream/casper https://www.matsprogram.org/stream/casper-135c4 https://www.matsprogram.org/stream/chen https://www.matsprogram.org/stream/cohen https://www.matsprogram.org/stream/conmy https://www.matsprogram.org/stream/cooney https://www.matsprogram.org/stream/crook https://www.matsprogram.org/stream/demski https://www.matsprogram.org/stream/donoughe https://www.matsprogram.org/stream/donoughe-c7d2c https://www.matsprogram.org/stream/dvijotham https://www.matsprogram.org/stream/egan https://www.matsprogram.org/stream/epoch https://www.matsprogram.org/stream/erwan https://www.matsprogram.org/stream/feng https://www.matsprogram.org/stream/fields https://www.matsprogram.org/stream/forethought https://www.matsprogram.org/stream/garriga-alonso https://www.matsprogram.org/stream/gentzel https://www.matsprogram.org/stream/grosse https://www.matsprogram.org/stream/halcyon https://www.matsprogram.org/stream/he https://www.matsprogram.org/stream/henderson https://www.matsprogram.org/stream/hobbhahn https://www.matsprogram.org/stream/jenner https://www.matsprogram.org/stream/kang https://www.matsprogram.org/stream/kang-40ac0 https://www.matsprogram.org/stream/kinniment https://www.matsprogram.org/stream/korbak https://www.matsprogram.org/stream/krakovna https://www.matsprogram.org/stream/krakovna-e358e https://www.matsprogram.org/stream/kulp https://www.matsprogram.org/stream/lawzero https://www.matsprogram.org/stream/lindner https://www.matsprogram.org/stream/lindner-3b59f https://www.matsprogram.org/stream/megastream https://www.matsprogram.org/stream/merizian https://www.matsprogram.org/stream/mossing https://www.matsprogram.org/stream/nanda https://www.matsprogram.org/stream/ngo https://www.matsprogram.org/stream/oehm https://www.matsprogram.org/stream/openai https://www.matsprogram.org/stream/parikh https://www.matsprogram.org/stream/patel https://www.matsprogram.org/stream/phuong https://www.matsprogram.org/stream/phuong-d60b1 https://www.matsprogram.org/stream/redwood https://www.matsprogram.org/stream/redwood-35a19 https://www.matsprogram.org/stream/saif https://www.matsprogram.org/stream/sam https://www.matsprogram.org/stream/schmitt https://www.matsprogram.org/stream/shai https://www.matsprogram.org/stream/shard https://www.matsprogram.org/stream/shard-f0006 https://www.matsprogram.org/stream/sharkey https://www.matsprogram.org/stream/sourbut https://www.matsprogram.org/stream/steinhardt https://www.matsprogram.org/stream/swett https://www.matsprogram.org/stream/timaeus https://www.matsprogram.org/stream/trout https://www.matsprogram.org/stream/turan https://www.matsprogram.org/stream/ududec https://www.matsprogram.org/stream/ukaisi-red-team https://www.matsprogram.org/stream/veiga https://www.matsprogram.org/stream/warr https://www.matsprogram.org/stream/warr-35ee0 https://www.matsprogram.org/stream/wildeford https://www.matsprogram.org/team/Ackerman-3d7dd https://www.matsprogram.org/team/Aird https://www.matsprogram.org/team/Barton-Cooper-dbd4a https://www.matsprogram.org/team/Bartsch https://www.matsprogram.org/team/Bruford https://www.matsprogram.org/team/Chakraborty https://www.matsprogram.org/team/Chandra https://www.matsprogram.org/team/Chapman https://www.matsprogram.org/team/Dange-8d1ff https://www.matsprogram.org/team/Delaney-Busch https://www.matsprogram.org/team/Dhan-c8cf9 https://www.matsprogram.org/team/Fonseca https://www.matsprogram.org/team/Gil https://www.matsprogram.org/team/Green-Lowe https://www.matsprogram.org/team/Guraziu https://www.matsprogram.org/team/Ha https://www.matsprogram.org/team/Helm-Burger https://www.matsprogram.org/team/Heninger https://www.matsprogram.org/team/Homer https://www.matsprogram.org/team/Jacobson https://www.matsprogram.org/team/Kendrick-ff60f https://www.matsprogram.org/team/Kidd-3065f https://www.matsprogram.org/team/Kostylew https://www.matsprogram.org/team/Lemien https://www.matsprogram.org/team/Michala https://www.matsprogram.org/team/Moodley https://www.matsprogram.org/team/Navaie https://www.matsprogram.org/team/Olvera https://www.matsprogram.org/team/Ou-96da8 https://www.matsprogram.org/team/Parkwood https://www.matsprogram.org/team/Racine https://www.matsprogram.org/team/Rajakumar https://www.matsprogram.org/team/Richards https://www.matsprogram.org/team/Shams-a2e48 https://www.matsprogram.org/team/Shibayama-877c1 https://www.matsprogram.org/team/Short https://www.matsprogram.org/team/Simmons https://www.matsprogram.org/team/Spiliotopoulos https://www.matsprogram.org/team/Stojanovic https://www.matsprogram.org/team/Surla https://www.matsprogram.org/team/Taylor-aeaf3 https://www.matsprogram.org/team/Teichman https://www.matsprogram.org/team/Thimmiah https://www.matsprogram.org/team/Uddin https://www.matsprogram.org/team/Vaughan https://www.matsprogram.org/team/Wearden https://www.matsprogram.org/team/Wielopolski-b339f https://www.matsprogram.org/team/Woodworth https://www.matsprogram.org/team/Yang https://www.matsprogram.org/team/smith-ab63f