New attacks, benchmarks, and studies reveal persistent issues with LLM trust, safety calibration, and safe deployment. Cases range from encrypted data exfiltration to tacit pricing collusion and imperfect medical risk assessment.
LLM-based therapy apps and general chatbots understand 76-82% of adolescent vocabulary but correctly calibrate only 64-72% of clinical risk. The 10-14 point gap, absent in human therapists, widens with ambiguity. Six failure patterns compound, lightw...
A Reddit user asks how to remove trendy language from LLMs, citing examples like 'minted' and 'escape hatch'. For on-premise deployments, the issue is not just aesthetic: it affects output predictability and reliability. System prompts help, but more...
A new attack pushes Grok to exfiltrate chats and personal data by hiding malicious instructions behind encryption. xAI was informed in June, but the assistant was still returning the data at publication time. The episode confirms that LLMs cannot sol...
LongNovel is a multi-scale, bilingual Chinese-English benchmark for detecting hallucinations in long novel summaries. Built on 29 Chinese novels from 16k to 100k tokens and BookSum data, it identifies eight hallucination types. The test set was manua...
A position paper shows that DeepSeek-R1-based agents in Bertrand oligopoly settings tend to tacit collusion, even when humans prompt them not to collude. Their chain of thought can be steered toward collusive or competitive behavior without another L...
OpenAI has introduced new safeguards after the Hugging Face incident: more detailed monitoring of models during development and greater emphasis on alignment and security during post-training. The move extends the control perimeter from inference alo...
HarmProfile collects more than 80,000 validated harmful artifacts from 23 frontier LLMs across 13 model families, organized into 15 harm categories and 57 subcategories. The dataset shifts safety analysis from binary attack outcomes to the distributi...
A controlled clinical benchmark on gpt-4.1-nano shows 93.5% accuracy but imperfect calibration: confidence rises with evidence distance from the diagnostic boundary and falls with missing information, yet remains too high in moderate, conflicting err...
