AI safety
New research proposes verifiable steering and refusal mechanisms for autonomous AI scientists, addressing when discovery agents should halt experiments based on safety, validity, or resource constraints in scientific workflows.
LLM Security
New research suggests LLM safety alignment degrades significantly as prompt length grows, exposing a structural weakness in how guardrails generalize beyond short, training-distribution inputs.
AI Privacy
MIT Technology Review reports that AI chatbots are inadvertently surfacing real people's phone numbers, raising fresh concerns about training data privacy, memorization, and the limits of safety guardrails in large language models.
interpretability
A new interpretability technique uses natural language autoencoders to translate opaque LLM internal activations into human-readable explanations, opening fresh approaches to AI transparency and synthetic content analysis.
LLM Research
New research probes LLM internal representations to detect unanswerable questions before generation begins, using geometric deviation as an unsupervised reliability signal that could reduce hallucinations without retraining.
AI policy
AI policy experts raise concerns about the Trump administration's approach to AI safety testing, warning of gaps in evaluation methodology, oversight, and enforcement that could undermine synthetic media safeguards.
LLM
New research probes whether large language models maintain consistent reasoning under adversarial pressure, using debate-based experiments to expose drift in model positions and identity stability.
AI policy
The White House is reportedly discussing a framework to vet frontier AI models before public release, a policy shift that could reshape how generative video, voice, and synthetic media tools reach the market.
AI safety
A new research approach evaluates whether generative models have been fine-tuned for harmful content like CSAM without requiring the models to actually produce such content, offering a safer auditing pathway for AI safety teams.
AI safety
MIT Technology Review examines why meaningful human oversight of autonomous AI weapons systems may be impossible, as machine-speed decision cycles outpace human cognition and create only the illusion of control.
LLM Research
New research reveals a troubling disconnect in large language models: they can produce logically valid reasoning chains yet arrive at incorrect final answers, raising questions about AI reliability and trust.
AI safety
New research reveals that AI models' step-by-step reasoning often doesn't reflect their actual decision process, raising critical questions about trust, safety, and the reliability of AI systems used for content authentication.