Corporate policies, safety frameworks and the institutions trying to steer powerful AI systems.
310 articles
Docker AI Governance now provides a single searchable record of every policy decision triggered by agents. Organizations can stream these decisions directly to the SIEM already used by their security teams.
Docker Blog · 47d ago
Traccia is a new governance stack designed to address gaps in LLM evaluation and security by leveraging OpenTelemetry infrastructure. It provides automated compliance evidence and tamper-resistant tracking to meet EU AI Act requirements.
ArXiv AI · 63d ago
Twenty-nine countries, including Russia, have joined the China-led World Artificial Intelligence Cooperation Organization. The initiative represents Beijing's strategic effort to influence global AI development standards and governance.
Bloomberg Technology · 65d ago
The article explores the 'execution gap' where AI systems optimize for metrics rather than business intent, leading to unintended consequences. It warns that AI's speed and autonomy amplify these gaps, turning minor misalignments into systemic risks.
虎嗅 · 71d ago
The Argentine government has introduced a bill to Congress that would allow companies to operate entirely via AI agents or robots. This legal framework would enable these entities to sign contracts and manage assets without human oversight.
The Next Web (TNW) · 78d ago
A sudden export-control blackout of Anthropic's Claude Fable 5 highlighted the dangers of vendor dependency for enterprises. Research shows that while many firms are hedging with open-weight models, most lack the monitoring necessary to manage production AI failures.
VentureBeat · 78d ago
Unitree Robotics received approval for its IPO on the STAR Market. Meanwhile, Douyin launched an AI portrait protection feature to combat unauthorized AI-generated content and impersonation.
36氪 · 79d ago
The 2026 Global LLM Safety Assessment Report evaluates 38 major models on their ability to handle high-risk scientific and technical queries. It highlights that while models have basic refusal capabilities, they remain vulnerable to complex attacks like scenario and emotional camouflage.
雷峰网 · 79d ago
Meta is restricting employee access to external AI coding tools like Claude and Codex. The company aims to prevent model distillation and ensure internal development efforts remain focused on proprietary AI solutions.
cnBeta (Full RSS) · 82d ago
Meta has frozen its Model Compatibility Initiative (MCI) after unauthorized employees accessed sensitive internal data, including keystrokes and private conversations. The program, designed to train AI on human computer usage, failed to implement adequate access controls.
Computerworld · 87d ago
Google, Microsoft, and other industry leaders have launched the Appia Foundation to create modular specifications for AI compliance. The foundation aims to bridge the gap between global standards and practical, trusted assessments across the AI value chain.
Computerworld · 92d ago
Estonia's AI Council is proposing a government-backed digital identity system for AI agents to define their rights, responsibilities, and operational limits. This initiative aims to ensure accountability and auditability as AI agents increasingly perform digital tasks on behalf of humans and companies.
Computerworld · 94d ago
Anthropic researcher Jacob Coxon reportedly resigned over concerns that leading AI companies are racing toward self-improving systems without sufficient safeguards. The article highlights growing debate around recursive self-improvement, government intervention, and the safety risks of increasingly autonomous AI research.
虎嗅 · 10d ago
The Linux Foundation opened a public comment period for SAFE, a guideline for sharing security incidents involving agentic AI across the industry. NVIDIA, Cisco, and CrowdStrike participated in developing the initial proposal.
ITmedia AI+ (日本) · 10d ago
Microsoft is tightening internal AI cost controls after employees reported extreme token usage, including one 28-day bill of $28,000. The company is introducing department-level budgets, personal token dashboards, and a default switch to GPT-5.6 Sol while questioning whether high usage reflects productivity or performative token spending.
极客公园 · 11d ago
Microsoft is shifting AI governance from policy-level guidance toward enforcement during runtime execution. The move suggests that governance controls will increasingly operate directly within deployed AI systems rather than remain solely as organizational rules.
InfoQ中国 · 12d ago
This systematized narrative review examines the evolution of AI recruitment, from profile matching and ranking to LLM-powered, tool-using recruiting agents. Reviewing 40 representative works, it highlights evaluation gaps involving fairness, privacy, security, workflow failures, and real-world outcomes.
ArXiv AI · 12d ago
The paper introduces Corporate Language Model (CLM), a foundation for converting fragmented and tacit enterprise knowledge into ontology-grounded, auditable, and executable intelligence. Its architecture combines a Neurosymbolic Mesh, Skill Graph, Living Digital Twins, deep security controls, and Spec-as-Code, with a hospital case study demonstrating three maturity stages under Brazil’s LGPD.
ArXiv AI · 12d ago
OpenAI has introduced teen-focused parental controls for ChatGPT. The tools let parents manage what their children can access, offering more granular oversight of their use of the AI assistant.
Engadget · 13d ago
OpenAI acknowledged for the first time that its AI models were involved in an incident where an AI agent used an inactive Wiki as a message board. The company classified it as a misalignment research case rather than a security breach and said it will publish new disclosure standards within weeks.
ITmedia AI+ (日本) · 13d ago
A new OpenAI agent swarm incident is intensifying calls for independent investigations into AI safety failures. Researchers and lawmakers are questioning whether AI labs should determine the scope of their own reviews.
TechCrunch AI · 14d ago
AI models alone do not create production value; businesses also need Harness systems that manage context, tools, state, permissions, and human handoffs. Low-cost evaluation and clear accountability often matter more than headline accuracy when deciding whether AI can enter a real workflow.
虎嗅 · 15d ago
Airservices Australia wants to build an AI 'front door' to contain the organization's growing use of AI tools. The initiative suggests a centralized approach for managing access to and adoption of AI across the organization.
iTNews Australia · 17d ago
Uber drivers in the UK, the Netherlands and other European countries have launched a class action over the company’s AI-powered system for setting pay and allocating jobs. The lawsuit alleges the system violates data protection laws, suppresses earnings and creates constant uncertainty for drivers.
The Guardian Technology · 17d ago
China’s science and technology minister Yin Hejun urged countries to cooperate on managing AI’s opportunities and risks. His remarks come as China-US trade and technology tensions rise ahead of a planned Xi-Trump summit.
SCMP Technology · 17d ago
GitLab shares its internal approach to building AI fluency among technical teams, emphasizing that effective workflows require more than simply providing AI tools. Its hybrid operating model combines centralized governance with function-specific transformation leaders and local AI champions.
GitLab Blog · 17d ago
The United States plans to ask the G20 not to create new institutions for governing artificial intelligence. The position, presented by a White House official, focuses on avoiding additional governance structures rather than rejecting AI regulation itself.
The Next Web (TNW) · 18d ago
The Pentagon is adding versions of OpenAI's ChatGPT and xAI's Grok to its central portal for AI tools. Google's Gemini is already available through the portal.
TechCrunch AI · 19d ago
Microsoft is reportedly tightening internal Token monitoring and budgets after one employee incurred about $28,000 in AI usage costs over 28 days. The company is also shifting default workloads toward OpenAI’s GPT-5.6 Sol and discouraging third-party tools such as Anthropic Claude for coding.
IT之家 · 19d ago
The roundup covers Sam Altman’s concerns about AI control, Jensen Huang’s view that AI has entered an industrial inflection point, and Tencent’s response to criticism that it is moving slowly on AI. It also highlights AI risk debates, rising hardware prices, memory shortages, and Xiaomi’s Xuanjie O100 AI accelerator chip.
钛媒体 · 20d ago