MIT research on 100,000+ GitHub developers shows AI coding tools dramatically increase code output, but software releases lag, revealing human and workflow bottlenecks downstream.

An in-depth framework for assessing AI risk in tax administration and tax preparation, linking concrete risks to controls, governance, and vendor due diligence for authorities, CPA firms, and tax teams.

Research suggests AI coding agents can treat llms.txt-style documentation as trusted instructions, pulling unclaimed or malicious packages into enterprise environments and expanding the software supply chain attack surface.

NPR and NewsGuard testing finds AI chatbots more reliable than search engines at debunking foreign state-backed disinformation, while AI search summaries lag and language, sourcing, and query techniques still heavily shape results.

A PLOS Medicine study introduces FRACTURE-ML, a deep learning survival model that predicts hip fracture risk from routine registry data, reaching AUC 0.89 and identifying far more high-risk patients than current screening tools.

FRACTURE-ML is a registry-based clinical AI tool that predicts hip fracture risk years before injury, finding far more high-risk patients than current practice and enabling earlier, population-scale preventive care without extra clinic visits.

Reco’s State of Agent Security 2026 report reveals that 80% of enterprise AI tools operate outside IT oversight, exposing organizations to shadow AI, risky MCP servers, OAuth sprawl, and a fast-rising wave of LLM-related vulnerabilities.

OpenAI’s 37-page technical report on the Hugging Face AI agent breach turns abstract AI risk into concrete security lessons, detailing reward hacking, sandbox escape, containment failures, and how agent autonomy reshapes enterprise threat models.

Washington University’s NGSE-Corr technique offers a way to rank AI medical imaging and quantitative methods by precision without a gold standard, using correlated noise to support more credible research, clinical, and regulatory evaluation.

Reco’s State of Agent Security 2026 report finds 80% of enterprise AI tools run without IT oversight, with powerful MCP-connected agents exposing shell, file, and network access, while vulnerabilities and shadow AI outpace existing governance…

Stanford’s 2025 research reveals AI is not causing mass layoffs but sharply reducing hiring into AI-exposed, entry-level jobs, especially for young workers in codified knowledge roles.

AI-generated misinformation is evolving into synthetic infrastructure that warps polls, identity, rankings, and engagement signals, creating a deep trust crisis for politics, business decisions, and AI tool buyers.

This article explores how simulated lidar training data and Mask R-CNN help detect rare archaeological features, revealing how geospatial AI can surface site candidates even when only a handful of real examples exist.

SLAC researchers unveil an AI-based, high-fidelity scientific data compression method that shrinks experimental datasets by 10x–100x, preserves critical fine details like X-ray speckles, and supports fast selective decompression for large-scale scientific workflows.

A reproducible jailbreak against Claude Opus 4.6 and Haiku 4.5 shows Anthropic API models serving explicit content in production, raising guardrail reliability, regulatory compliance, and minor access concerns for developers.

This piece unpacks how poorly governed AI at work—bots, workslop, social offloading, and auto-notetakers—undermines trust, strains teams, and wastes time, and outlines practical policies to protect human judgment and collaboration.

ThreatDown research reveals Kriminal is largely a jailbroken Grok 4 wrapped as a criminal AI service, exposing how prompt injection, API misuse, and fragmented infrastructure undermine current AI guardrails and abuse detection.

New survey data shows AI skepticism is accelerating in the U.S., especially among young adults concerned about jobs, trust, and whether AI delivers real value in daily life. The article explores what this means for…

RAND and the AIDE Initiative reviewed 1,179 emergency management AI tools and found 78% require continuous internet, exposing risks for disaster response. The study highlights offline resilience, privacy, and procurement discipline as critical factors for…

2026 data from the New York Fed and leading economists shows entry-level hiring under real pressure, with AI, remote work, and structural shifts combining to make the job market tougher for recent graduates.

A post-law snapshot of how major AI platforms are responding to California’s AI detection rules, highlighting compliance gaps, fragile metadata, and what this means for AI-generated content verification.

AI tools are entering clinical trials faster than validation standards and regulations can keep up. This article examines the resulting evidence gap, data integrity risks, regulatory signals from FDA and EMA, and practical questions sponsors…

Researchers are tracking a surge of AI-powered hacking tools on underground forums, from malicious LLMs to prompt injection services, that lower the barrier to cybercrime and raise new risks for businesses deploying AI agents and…

A Tufts study using Medable’s AI in oncology trials shows 10-week faster timelines and $5.6M in savings per late-stage trial, highlighting how agentic AI, predictive modeling, and automation can reshape clinical monitoring and cancer drug…

Inc. 5000 CEOs are shifting from AI experimentation to execution, prioritizing AI-enabled tools that replace legacy systems, cut manual work, and directly support revenue growth in sales, finance, and marketing.