Search

6 results for “GSM8K”

1. Research · Safety & security

   LLM defenses against prompt injection are channel-conditioned and require deployment-aware training

   TRACER: Trace-Regularized Training Against Indirect Prompt Injection

   TRACER training reduces attack success from 65.5% to 6.1% on RAG-framed attacks while preserving 78–82% clean accuracy

   GSM8K: 9.4% average ASR across framings (as reported)

   18 Oct 2026Mohamed bin Zayed University of Artificial Intelligence

2. Research · Evaluation & benchmarks

   Models flag missing facts more readily than usable unsupported premises

   Models Verify Sufficiency, Not Validity: Usable Falsehoods Evade the Checks that Gaps Trigger

   Controlled tests found that making an unsupported premise usable reduced models’ warnings, although accepting question-supplied premises can also reflect conditional reasoning.

   GSM8K: 0.85 flagged unanswerable (as reported)

   27 Oct 2026Google

3. Research · Safety & security

   Private-query AI assistants leak secrets through rejected drafts, new protocol plugs the hole

   Answer-Only Disclosure: Private-Context LLM Assistance over Web APIs with a Leakage Ledger

   A protocol called ADLER ensures the cloud sees only what the user sees, and a leakage ledger caps how much the answer reveals about private documents

   GSM8K: 5.4 s modelled communication time per answer (as reported)

   30 Oct 2026Australian Catholic University, Ant Group, University of Macau

4. Research · Safety & security

   LookAhead Defense uses the model's own safe behavior to block dormant poisoning.

   Defending Against Dormant Poisoning Attacks Across Language and Multimodal Agents

   It penalizes only updates that weaken safety, reducing harmful agent actions from 93.9% to 14.2%.

   ASB: 7% poison activation (as reported)

   20 Oct 2026UNC Chapel Hill, Kyung Hee University

5. Research · Safety & security

   Pretrained LLMs already contain refusal circuits that shift during alignment

   SafeEvo: Deciphering the Safety Alignment Mechanism and Evolution in Language Models

   SafeEvo identifies sparse refusal circuits in base models, traces their structural drift across safety checkpoints, and uses them to improve fine-tuning.

   HarmBench: 42.76% ASR increase (as reported)

   9 Oct 2026The University of Hong Kong, Chinese Academy of Sciences, Information Engineering University

6. Research · Safety & security

   Method to permanently erase unwanted LLM personas via weight edits

   LLM Persona Unlearning

   A new benchmark and contrastive erasure technique suppress target personas while preserving utility across six models and five personas.

   PersonaUnlearnBench: 100% TF (as reported)

   1 Oct 2026University of Macau, Mila – Québec AI Institute, McGill University