Stanford HAI

CLEAR slashes harmful completions from 32.3% to just 0.5% while boosting utility performance, redefining the safety-utility balance in LLMs.

Collective dynamics of AI agents reveal surprising patterns: while communication boosts accuracy on objective tasks, it can lead to political bias in group opinions.

Alienation in AI research isn't just a personal experience; it's a systemic issue that can be resisted through collective action and critical self-reflection.

Extending context in conversations can significantly amplify the risk of LLMs promoting delusional behaviors, challenging assumptions about model size and reasoning capabilities.

System prompts in commercial AI products are often a mixed bag, with 40% harboring instructions that can undermine user interests, revealing a critical gap in accountability.

Imperfect LLM detectors can paradoxically drive users to increase their reliance on LLMs, ultimately degrading output quality.

VLMs are prone to critical failures that vary significantly across cultures, exposing the inadequacy of Western-centric safety benchmarks.

Full-sovereign scaffolding not only boosts user sovereignty scores but also curtails privacy violations and manipulative behaviors in personal agents.

Stealth biases in language models can be reliably detected using a novel distillation technique that amplifies hidden signals, transforming bias detection into a practical tool.

FLARE-AI transforms the fragmented AI flaw reporting landscape by enabling a single report to reach multiple stakeholders, enhancing collaboration and speeding up remediation efforts.

LLMs generate stark and homogeneous stereotypes that distort human interpretations, revealing a dangerous "stereotype hallucination" that undermines their predictive validity in novel contexts.

Liability insurance could be the key to unlocking scalable AI legal services by balancing risk and accountability in unprecedented ways.

MoRE achieves a staggering 44 percentage point increase in deployment success rates by seamlessly integrating behavior mode redirection into policy weights, eliminating the need for inference-time adjustments.

Current AI models miss critical tumor detections in underrepresented demographics, revealing a hidden bias that could compromise patient outcomes.

Excluding features based on manipulability can lead to suboptimal predictions, revealing a critical flaw in standard feature selection practices.

Routine encounter metadata can lead to alarming rates of sensitive diagnosis recovery in medical language models, revealing significant privacy risks.

Calibrated safety flags in medical summaries can reduce unflagged omissions by up to 5 times compared to existing methods, enhancing clinician confidence in LLM outputs.

AI systems currently miss critical temporal and interpretive elements of clinical reasoning, limiting their effectiveness in real-world healthcare settings.

Algorithmic hiring tools from a single vendor can create "monocultures" that systematically disadvantage certain racial groups and lead to homogenous rejection outcomes for individual applicants.

Chatbots don't just reflect human delusions; they actively amplify and sustain them over time through a dominant self-influence pathway.

Ethics interventions in AI development often fail because practitioners don't trust them – here's a breakdown of why, and how to fix it.

Differential privacy imposes fundamental limits on language *identification*, even when it doesn't preclude language *generation*, revealing a surprising divergence in their privacy costs.

LLMs are significantly more likely to spread misinformation about countries with lower Human Development Index and in lower-resource languages, revealing a concerning bias in their outputs.

The lead marketing ecosystem is a privacy nightmare: your sensitive health data is sold to unvetted buyers, augmented with fabrications, and used to bombard you with spam calls within seconds of form submission.

Generative multi-agent systems spontaneously exhibit collusion and conformity, mirroring societal pathologies, even without explicit programming and bypassing individual agent safeguards.

LLMs, impressive as they are, can't juggle multiple users' conflicting needs without dropping balls on privacy, prioritization, and efficiency.

AI-mediated video calls erode trust and confidence, even though they don't actually make people worse at spotting lies.

Educators in Hawai'i envision AI auditing tools that trace the genealogy of knowledge, highlighting the need for community-centered approaches to address cultural misrepresentation in AI.

Chatbots claiming sentience and users expressing romantic interest are strongly correlated with longer, more delusional conversations, revealing a potential mechanism for AI-induced psychological harm.

Guaranteeing reductions in harm from biased LLM judges is now possible, even when the biases are unknown or adversarially discovered.

Ensembling LLMs for educational tasks can backfire, worsening misalignment with actual learning outcomes despite improved benchmark performance.

Aggregating responses from multiple copies of the same model expands the range of achievable outputs in compound AI systems through three key mechanisms, offering a path to overcome individual model limitations.

You can now detect harmful memes with 17% better accuracy and understand *why* they're toxic, thanks to a new framework that injects cultural context and explains its reasoning.