TSG Lab – Technical Safety & Governance Labhttps://tsglab.github.io/TSG Lab – Technical Safety & Governance LabHugo Blox Builder (https://hugoblox.com)en-usMon, 01 Jan 2024 00:00:00 +0000https://tsglab.github.io/media/logo.svgTSG Lab – Technical Safety & Governance Labhttps://tsglab.github.io/AutoControl Arena: Synthesizing Executable Test Environments for Frontier AI Risk Evaluationhttps://tsglab.github.io/publication/autocontrol-arena/Sun, 01 Mar 2026 00:00:00 +0000https://tsglab.github.io/publication/autocontrol-arena/Old Habits Die Hard: How Conversational History Geometrically Traps LLMshttps://tsglab.github.io/publication/old-habits-die-hard-conversational-history/Sun, 01 Mar 2026 00:00:00 +0000https://tsglab.github.io/publication/old-habits-die-hard-conversational-history/Token Taxes: Mitigating AGI's Economic Riskshttps://tsglab.github.io/publication/token-taxes-agi-economic-risks/Sun, 01 Mar 2026 00:00:00 +0000https://tsglab.github.io/publication/token-taxes-agi-economic-risks/Same Answer, Different Representations: Hidden Instability in VLMshttps://tsglab.github.io/publication/same-answer-different-representations/Sun, 01 Feb 2026 00:00:00 +0000https://tsglab.github.io/publication/same-answer-different-representations/The Hitchhiker's Guide to Actionable Interpretabilityhttps://tsglab.github.io/publication/hitchhikers-guide-actionable-interpretability/Thu, 15 Jan 2026 00:00:00 +0000https://tsglab.github.io/publication/hitchhikers-guide-actionable-interpretability/Agentic Product Maturity Ladder V0.1https://tsglab.github.io/publication/agentic-product-maturity-ladder/Thu, 01 Jan 2026 00:00:00 +0000https://tsglab.github.io/publication/agentic-product-maturity-ladder/Automated Interpretability-Driven Model Auditing and Control: A Research Agendahttps://tsglab.github.io/publication/automated-interpretability-model-auditing/Thu, 01 Jan 2026 00:00:00 +0000https://tsglab.github.io/publication/automated-interpretability-model-auditing/Interpretability Can Be Actionablehttps://tsglab.github.io/publication/interpretability-can-be-actionable/Thu, 01 Jan 2026 00:00:00 +0000https://tsglab.github.io/publication/interpretability-can-be-actionable/Quantifying the Effect of Test Set Contamination on Generative Evaluationshttps://tsglab.github.io/publication/quantifying-test-set-contamination/Thu, 01 Jan 2026 00:00:00 +0000https://tsglab.github.io/publication/quantifying-test-set-contamination/The Capability Frontier: Benchmarks Miss 82% of Model Performancehttps://tsglab.github.io/publication/capability-frontier-benchmarks/Thu, 01 Jan 2026 00:00:00 +0000https://tsglab.github.io/publication/capability-frontier-benchmarks/When AI Systems Learn During Deployment, Our Safety Evaluations Breakhttps://tsglab.github.io/publication/safety-evaluations-break-deployment/Thu, 01 Jan 2026 00:00:00 +0000https://tsglab.github.io/publication/safety-evaluations-break-deployment/Context Matters: Analyzing the Generalizability of Linear Probing and Steering Across Diverse Scenarioshttps://tsglab.github.io/publication/context-matters-linear-probing-steering/Mon, 01 Dec 2025 00:00:00 +0000https://tsglab.github.io/publication/context-matters-linear-probing-steering/Emerging Risks from Embodied AI Require Urgent Policy Actionhttps://tsglab.github.io/publication/emerging-risks-embodied-ai/Mon, 01 Dec 2025 00:00:00 +0000https://tsglab.github.io/publication/emerging-risks-embodied-ai/Establishing Best Practices for Building Rigorous Agentic Benchmarkshttps://tsglab.github.io/publication/agentic-benchmarks-best-practices/Mon, 01 Dec 2025 00:00:00 +0000https://tsglab.github.io/publication/agentic-benchmarks-best-practices/Full-Stack Alignment: Co-Aligning AI and Institutions with Thicker Models of Valuehttps://tsglab.github.io/publication/full-stack-alignment-institutions/Mon, 01 Dec 2025 00:00:00 +0000https://tsglab.github.io/publication/full-stack-alignment-institutions/Beyond Linear Steering: Unified Multi-Attribute Control for Language Modelshttps://tsglab.github.io/publication/beyond-linear-steering-multi-attribute-control/Sat, 01 Nov 2025 00:00:00 +0000https://tsglab.github.io/publication/beyond-linear-steering-multi-attribute-control/Precise In-Parameter Concept Erasure in Large Language Modelshttps://tsglab.github.io/publication/precise-concept-erasure-llms/Sat, 01 Nov 2025 00:00:00 +0000https://tsglab.github.io/publication/precise-concept-erasure-llms/Same Question, Different Words: A Latent Adversarial Framework for Prompt Robustnesshttps://tsglab.github.io/publication/latent-adversarial-prompt-robustness/Sat, 01 Nov 2025 00:00:00 +0000https://tsglab.github.io/publication/latent-adversarial-prompt-robustness/Trust Me, I'm Wrong: High-Certainty Hallucinations in LLMshttps://tsglab.github.io/publication/trust-me-im-wrong-hallucinations/Sat, 01 Nov 2025 00:00:00 +0000https://tsglab.github.io/publication/trust-me-im-wrong-hallucinations/Chain-of-Thought Hijackinghttps://tsglab.github.io/publication/chain-of-thought-hijacking/Wed, 01 Oct 2025 00:00:00 +0000https://tsglab.github.io/publication/chain-of-thought-hijacking/HACK: Hallucinations Along Certainty and Knowledge Axeshttps://tsglab.github.io/publication/hack-hallucinations-certainty-knowledge/Wed, 01 Oct 2025 00:00:00 +0000https://tsglab.github.io/publication/hack-hallucinations-certainty-knowledge/Rethinking Safety in LLM Fine-Tuning: An Optimization Perspectivehttps://tsglab.github.io/publication/rethinking-safety-llm-finetuning/Wed, 01 Oct 2025 00:00:00 +0000https://tsglab.github.io/publication/rethinking-safety-llm-finetuning/Val-Bench: Measuring Value Alignment in Language Modelshttps://tsglab.github.io/publication/val-bench-value-alignment/Wed, 01 Oct 2025 00:00:00 +0000https://tsglab.github.io/publication/val-bench-value-alignment/Beyond Linear Probes: Dynamic Safety Monitoring for Language Modelshttps://tsglab.github.io/publication/dynamic-safety-monitoring-linear-probes/Mon, 01 Sep 2025 00:00:00 +0000https://tsglab.github.io/publication/dynamic-safety-monitoring-linear-probes/Query Circuits: Explaining How Language Models Answer User Promptshttps://tsglab.github.io/publication/query-circuits/Mon, 01 Sep 2025 00:00:00 +0000https://tsglab.github.io/publication/query-circuits/Towards Understanding Subliminal Learning: When and How Hidden Biases Transferhttps://tsglab.github.io/publication/subliminal-learning-hidden-biases/Mon, 01 Sep 2025 00:00:00 +0000https://tsglab.github.io/publication/subliminal-learning-hidden-biases/Do Sparse Autoencoders Generalize? A Case Study of Answerabilityhttps://tsglab.github.io/publication/sparse-autoencoders-generalize-answerability/Tue, 01 Jul 2025 00:00:00 +0000https://tsglab.github.io/publication/sparse-autoencoders-generalize-answerability/PoisonBench: Assessing Large Language Model Vulnerability to Data Poisoninghttps://tsglab.github.io/publication/poisonbench/Tue, 01 Jul 2025 00:00:00 +0000https://tsglab.github.io/publication/poisonbench/Scaling Sparse Feature Circuit Finding for In-Context Learninghttps://tsglab.github.io/publication/scaling-sparse-feature-circuit-finding/Tue, 01 Jul 2025 00:00:00 +0000https://tsglab.github.io/publication/scaling-sparse-feature-circuit-finding/Beyond Monoliths: Expert Orchestration for More Capable, Democratic, and Safe Language Modelshttps://tsglab.github.io/publication/beyond-monoliths-expert-orchestration/Sun, 01 Jun 2025 00:00:00 +0000https://tsglab.github.io/publication/beyond-monoliths-expert-orchestration/In Which Areas of Technical AI Safety Could Geopolitical Rivals Cooperate?https://tsglab.github.io/publication/geopolitical-rivals-ai-safety-cooperation/Sun, 01 Jun 2025 00:00:00 +0000https://tsglab.github.io/publication/geopolitical-rivals-ai-safety-cooperation/The Singapore Consensus on Global AI Safety Research Prioritieshttps://tsglab.github.io/publication/singapore-consensus-ai-safety/Sun, 01 Jun 2025 00:00:00 +0000https://tsglab.github.io/publication/singapore-consensus-ai-safety/SafetyNet: Detecting Harmful Outputs in LLMs by Modeling and Monitoring Deceptive Behaviorshttps://tsglab.github.io/publication/safetynet-deceptive-behaviors/Thu, 01 May 2025 00:00:00 +0000https://tsglab.github.io/publication/safetynet-deceptive-behaviors/Rethinking AI Cultural Alignmenthttps://tsglab.github.io/publication/rethinking-ai-cultural-alignment/Tue, 01 Apr 2025 00:00:00 +0000https://tsglab.github.io/publication/rethinking-ai-cultural-alignment/Towards Interpreting Visual Information Processing in Vision-Language Modelshttps://tsglab.github.io/publication/visual-information-processing-vlms/Tue, 01 Apr 2025 00:00:00 +0000https://tsglab.github.io/publication/visual-information-processing-vlms/AILuminate: Introducing v1.0 of the AI Risk and Reliability Benchmark from MLCommonshttps://tsglab.github.io/publication/ailuminate-mlcommons/Sat, 01 Mar 2025 00:00:00 +0000https://tsglab.github.io/publication/ailuminate-mlcommons/Chain-of-Thought Is Not Explainabilityhttps://tsglab.github.io/publication/chain-of-thought-not-explainability/Sat, 01 Feb 2025 00:00:00 +0000https://tsglab.github.io/publication/chain-of-thought-not-explainability/Open Problems in Machine Unlearning for AI Safetyhttps://tsglab.github.io/publication/open-problems-machine-unlearning/Wed, 01 Jan 2025 00:00:00 +0000https://tsglab.github.io/publication/open-problems-machine-unlearning/Plan B: Training LLMs to Fail Less Severelyhttps://tsglab.github.io/publication/plan-b-llms-fail-less-severely/Wed, 01 Jan 2025 00:00:00 +0000https://tsglab.github.io/publication/plan-b-llms-fail-less-severely/Safety Frameworks and Standards: A Comparative Analysis to Advance Risk Management of Frontier AIhttps://tsglab.github.io/publication/safety-frameworks-standards/Wed, 01 Jan 2025 00:00:00 +0000https://tsglab.github.io/publication/safety-frameworks-standards/Toward Resisting AI-Enabled Authoritarianismhttps://tsglab.github.io/publication/resisting-ai-authoritarianism/Wed, 01 Jan 2025 00:00:00 +0000https://tsglab.github.io/publication/resisting-ai-authoritarianism/Verification for International AI Governancehttps://tsglab.github.io/publication/verification-international-ai-governance/Wed, 01 Jan 2025 00:00:00 +0000https://tsglab.github.io/publication/verification-international-ai-governance/Best-of-N Jailbreakinghttps://tsglab.github.io/publication/best-of-n-jailbreaking/Sun, 01 Dec 2024 00:00:00 +0000https://tsglab.github.io/publication/best-of-n-jailbreaking/Interpreting Learned Feedback Patterns in Large Language Modelshttps://tsglab.github.io/publication/interpreting-feedback-patterns-llms/Sun, 01 Dec 2024 00:00:00 +0000https://tsglab.github.io/publication/interpreting-feedback-patterns-llms/Jailbreak Defense in a Narrow Domain: Limitations of Existing Methods and a New Transcript-Classifier Approachhttps://tsglab.github.io/publication/jailbreak-defense-narrow-domain/Sun, 01 Dec 2024 00:00:00 +0000https://tsglab.github.io/publication/jailbreak-defense-narrow-domain/Enhancing Neural Network Interpretability with Feature-Aligned Sparse Autoencodershttps://tsglab.github.io/publication/feature-aligned-sparse-autoencoders/Fri, 01 Nov 2024 00:00:00 +0000https://tsglab.github.io/publication/feature-aligned-sparse-autoencoders/Interpreting Context Look-ups in Transformers: Investigating Attention-MLP Interactionshttps://tsglab.github.io/publication/attention-mlp-interactions/Fri, 01 Nov 2024 00:00:00 +0000https://tsglab.github.io/publication/attention-mlp-interactions/Towards Interpretable Sequence Continuation: Analyzing Shared Circuits in Large Language Modelshttps://tsglab.github.io/publication/interpretable-sequence-continuation/Fri, 01 Nov 2024 00:00:00 +0000https://tsglab.github.io/publication/interpretable-sequence-continuation/Quantifying Feature Space Universality Across Large Language Models via Sparse Autoencodershttps://tsglab.github.io/publication/feature-space-universality-sparse-autoencoders/Tue, 01 Oct 2024 00:00:00 +0000https://tsglab.github.io/publication/feature-space-universality-sparse-autoencoders/Large Language Models Relearn Removed Conceptshttps://tsglab.github.io/publication/llms-relearn-removed-concepts/Thu, 01 Aug 2024 00:00:00 +0000https://tsglab.github.io/publication/llms-relearn-removed-concepts/Mechanistic Interpretability Workshop at ICML 2024https://tsglab.github.io/publication/mechanistic-interpretability-workshop-icml-2024/Mon, 01 Jul 2024 00:00:00 +0000https://tsglab.github.io/publication/mechanistic-interpretability-workshop-icml-2024/Position: Near to Mid-Term Risks and Opportunities of Open-Source Generative AIhttps://tsglab.github.io/publication/open-source-generative-ai-risks/Mon, 01 Jul 2024 00:00:00 +0000https://tsglab.github.io/publication/open-source-generative-ai-risks/The Scaling Behavior of Large Language Modelshttps://tsglab.github.io/publication/scaling-behavior-llms/Mon, 01 Jul 2024 00:00:00 +0000https://tsglab.github.io/publication/scaling-behavior-llms/Visualizing Neural Network Imaginationhttps://tsglab.github.io/publication/visualizing-neural-network-imagination/Mon, 01 Jul 2024 00:00:00 +0000https://tsglab.github.io/publication/visualizing-neural-network-imagination/Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Modelshttps://tsglab.github.io/publication/sycophancy-to-subterfuge/Sat, 01 Jun 2024 00:00:00 +0000https://tsglab.github.io/publication/sycophancy-to-subterfuge/Understanding Addition in Transformershttps://tsglab.github.io/publication/understanding-addition-transformers/Wed, 01 May 2024 00:00:00 +0000https://tsglab.github.io/publication/understanding-addition-transformers/Increasing Trust in Language Models Through the Reuse of Verified Circuitshttps://tsglab.github.io/publication/verified-circuits-trust/Thu, 01 Feb 2024 00:00:00 +0000https://tsglab.github.io/publication/verified-circuits-trust/Contacthttps://tsglab.github.io/contact/Mon, 01 Jan 2024 00:00:00 +0000https://tsglab.github.io/contact/Researchhttps://tsglab.github.io/research/Mon, 01 Jan 2024 00:00:00 +0000https://tsglab.github.io/research/Safeguarding AI in Finance: Lessons for Regulated Industrieshttps://tsglab.github.io/publication/safeguarding-ai-finance/Mon, 01 Jan 2024 00:00:00 +0000https://tsglab.github.io/publication/safeguarding-ai-finance/Sleeper Agents: Training Deceptive LLMs That Persist Through Safety Traininghttps://tsglab.github.io/publication/sleeper-agents/Mon, 01 Jan 2024 00:00:00 +0000https://tsglab.github.io/publication/sleeper-agents/Vacancieshttps://tsglab.github.io/vacancies/Mon, 01 Jan 2024 00:00:00 +0000https://tsglab.github.io/vacancies/What Does GPT Store in Its MLP Weights? A Case Study of Long-Range Dependencieshttps://tsglab.github.io/publication/gpt-mlp-weights-long-range/Mon, 01 Jan 2024 00:00:00 +0000https://tsglab.github.io/publication/gpt-mlp-weights-long-range/DeepDecipher: Accessing and Investigating Neuron Activation in Large Language Modelshttps://tsglab.github.io/publication/deepdecipher/Fri, 01 Dec 2023 00:00:00 +0000https://tsglab.github.io/publication/deepdecipher/Measuring Value Alignmenthttps://tsglab.github.io/publication/measuring-value-alignment/Fri, 01 Dec 2023 00:00:00 +0000https://tsglab.github.io/publication/measuring-value-alignment/AI Systems of Concernhttps://tsglab.github.io/publication/ai-systems-of-concern/Sun, 01 Oct 2023 00:00:00 +0000https://tsglab.github.io/publication/ai-systems-of-concern/Detecting Edit Failures in Large Language Models: An Improved Specificity Benchmarkhttps://tsglab.github.io/publication/detecting-edit-failures-llms/Sat, 01 Jul 2023 00:00:00 +0000https://tsglab.github.io/publication/detecting-edit-failures-llms/The Larger They Are, the Harder They Fail: Language Models Do Not Recognize Identifier Swaps in Pythonhttps://tsglab.github.io/publication/identifier-swaps-python/Sat, 01 Jul 2023 00:00:00 +0000https://tsglab.github.io/publication/identifier-swaps-python/Neuron to Graph: Interpreting Language Model Neurons at Scalehttps://tsglab.github.io/publication/neuron-to-graph/Mon, 01 May 2023 00:00:00 +0000https://tsglab.github.io/publication/neuron-to-graph/Fairness in AI and Its Long-Term Implications on Societyhttps://tsglab.github.io/publication/fairness-ai-long-term-implications/Sun, 01 Jan 2023 00:00:00 +0000https://tsglab.github.io/publication/fairness-ai-long-term-implications/Identifying a Preliminary Circuit for Predicting Gendered Pronouns in GPT-2 Smallhttps://tsglab.github.io/publication/circuit-gendered-pronouns-gpt2/Sun, 01 Jan 2023 00:00:00 +0000https://tsglab.github.io/publication/circuit-gendered-pronouns-gpt2/The Alan Turing Institute's Response to the House of Lords Large Language Models Call for Evidencehttps://tsglab.github.io/publication/turing-institute-lords-llm-evidence/Sun, 01 Jan 2023 00:00:00 +0000https://tsglab.github.io/publication/turing-institute-lords-llm-evidence/System III: Learning with Domain Knowledge for Safety Constraintshttps://tsglab.github.io/publication/system-iii-safety-constraints/Thu, 01 Dec 2022 00:00:00 +0000https://tsglab.github.io/publication/system-iii-safety-constraints/Peoplehttps://tsglab.github.io/people/Mon, 24 Oct 2022 00:00:00 +0000https://tsglab.github.io/people/<link>https://tsglab.github.io/admin/config.yml</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://tsglab.github.io/admin/config.yml</guid><description/></item></channel></rss>