Constitutional AI & AI Ethics

Achieving robust and unbiased unlearning in diffusion models is now possible with AutoAnchor, which enhances performance by over 30% without manual bias.

Despite reducing persona collapse by 80%, LLMs still struggle to match human adaptability in advice-giving, with users favoring the default persona even in challenging situations.

GRAM achieves a 5x reduction in training costs while allowing for precise control over AI capabilities, making it a game-changer for safe AI deployment.

Self-analysis of LLMs reveals that a stronger model ensemble can recover 86% of safety principles, challenging assumptions about tool reliability in safety-critical applications.

AMALIA's performance reveals that high agreement with human coders does not guarantee valid measurement of complex constructs, highlighting critical flaws in national language model evaluations.

Trustworthy agentic AI evaluation in decentralized energy markets hinges on balancing market utility and safety, revealing critical vulnerabilities in reward-maximizing strategies.

Mediation can withstand sophisticated adversarial attacks, maintaining market stability even when honest-agent utility is compromised.

Users who depend on manual context attachment experience a dramatic drop in task success, revealing a critical divide in AI utility that could reshape how we design intelligent systems.

Models can misrepresent unanswerable questions, but a new calibrated policy allows for precise control over when they should answer.

Communication style can dramatically shift triage outcomes, highlighting the risks of deploying chatbots trained on idealized patient interactions.

Psychological competence in AI evaluation could redefine how we assess the impact of AI on human cognition and decision-making.

Achieving up to 94.93% F1 scores, this innovative firewall architecture offers a robust solution for protecting sensitive data in LLM interactions.

AI-teacher collaboration outperforms imitation learning, boosting student performance by nearly 50% on challenging coding tasks.

Invisible perturbations can lead to a staggering 75.8% information loss in agentic crawlers without altering the human-visible content.

TokenWall slashes the attack success rate to 12.5% while ensuring a 97.4% pass rate for benign interactions, all with just 0.69 seconds of added latency.

Coded offloading not only boosts performance but also reveals a surprising delay-energy-privacy trade-off that challenges conventional task execution strategies.

Analysts trust AI predictions but still rely heavily on traditional methods, highlighting a critical gap in AI integration for crime linkage analysis.

T2I models overwhelmingly depict disability through the lens of stereotypes, with wheelchair imagery dominating representations of mobility impairment.

Non-sentient AI could be recognized as persons with moral status, challenging the prevailing belief that sentience is a prerequisite for personhood.

Training LLMs on the PLURAL dataset can reduce cultural misalignment by nearly 28%, making them more representative of global values.

AegisDx captures 78% of critical "must-not-miss" diagnoses, significantly outperforming traditional LLMs in both accuracy and safety.

The probability of a model trajectory entering unsafe regions can be exponentially small, but the geometry of the unsafe set critically influences how quickly training stabilizes.

A unified detection framework reveals that accurately estimating covariance can significantly enhance the identification of AI-generated artifacts across multiple domains.

Penalizing the decision-making path while rewarding the outcome can drastically reduce operational violations in real-world agent interactions.

Transforming data systems from passive repositories into active agents could redefine the landscape of autonomous automation and its safety protocols.

Reasoning inconsistencies in AI outputs are not just common; they can be systematically detected and vary significantly across models and tasks.

AI agents that understand social norms can outperform human-human interactions, achieving a 43% improvement in coordination.

Deployment rules can shift multi-agent AI outcomes dramatically, with fatality rates varying by up to 58 percentage points based solely on the chosen rule.

The orchestration layer can slash AI task costs by over 40% without sacrificing quality, fundamentally reshaping how enterprises approach agentic AI deployment.

Bias detection in neural networks can be revolutionized by analyzing latent spaces and activations, revealing how biases are embedded in model architecture itself.

User identity can dramatically shift LLM moral evaluations, revealing a troubling layer of contextual conditioning in AI responses.

A biased judge can silently disable skill retirement in self-evolving agents, leading to unnoticed performance degradation that can jeopardize deployment.

The unique challenges posed by agentic AI demand a fresh governance approach, as traditional frameworks may no longer suffice.

Shifts in attention to low-credibility content can erode societal trust, making credible information increasingly difficult to discern and correct.

AI governance frameworks fall short in ensuring operational resilience, necessitating a new approach to manage dependencies and risks effectively.

Fairness interventions can significantly mitigate the equity losses caused by differential privacy, especially when applied in post-processing stages.

Silent policy violations in tool-using LLMs can be mitigated by deterministic gates, improving success rates by over 12 percentage points in critical tasks.

Post-trip safety measures could drastically reduce risks for ride-hailing passengers in high-crime areas, addressing a critical gap in current safety protocols.

Cybersecurity AI agents expose the EU Cyber Resilience Act's fatal flaws, revealing that static security certifications are doomed in a landscape where vulnerabilities can be exploited before they are even known.

Personalization in AI-driven development can introduce significant biases, with age and gender influencing the very structure of generated code.

Operational reframing emerges as a critical risk signal, revealing that compliance can vary significantly across models and scenarios, challenging the notion of stable safety metrics in multi-agent LLMs.

Asymmetric reward design in deep reinforcement learning can drastically reduce false negatives in ransomware detection, achieving a remarkable 67.6% improvement over traditional methods.

Curvature-aware evaluation reveals hidden biases in 3D face reconstruction, challenging the fairness of existing models across diverse populations.

Robots may be acting without consent, highlighting a crucial gap in safety protocols that could lead to unintended social consequences.

Combining DRL with MPC not only enhances safety in exploration but also ensures stable policy convergence in complex physical systems.

Achieving trajectory-level differential privacy in adaptive streaming contexts without sacrificing performance is now feasible through an auditable buffering-aggregation approach.

Distinct manipulation profiles for major fraud types were uncovered, revealing significant gaps in actionable victim narrative details that AI can help bridge.

Static safety policies fail in offensive security, with ScopeJudge revealing that context-aware monitoring is crucial to avoid costly violations.

Many text-to-image models are safer than expected, but a subset poses significant risks that traditional evaluation methods fail to capture.

Trustworthy code generation can be achieved without post-generation adjustments, improving security and functionality simultaneously.