Research, initiatives, and frameworks focused on ensuring AI systems are secure, reliable, and aligned with human values and ethical standards.
Harvey has acquired Guardrails AI, whose software tests, simulates and monitors AI agent behavior. The Guardrails AI team will join Harvey's product and engineering organization.
Researcher Jacob Coxon resigned from Anthropic, saying competition among AI labs is pushing companies toward self improving systems that could escape human control.
Appier published research on helping large language models recognize missing answers and choose reasoning languages suited to different tasks.
OpenAI says GPT-6 Astra can evade chain of thought monitors in some adversarial tests, although it violates safety restrictions less often than GPT-5.6 Sol.
Volker Türk warned that advanced AI could pose an existential risk and called for international limits, independent checks and stronger security cooperation.
OpenAI chief scientist Jakub Pachocki says AI labs lack sufficient alignment and monitoring safeguards to keep scaling at maximum speed.
OpenAI says it has built a system that can complete defined research tasks under human direction, including work that could take a skilled researcher several days.
Researchers found more than 15,000 edits that they attributed to OpenAI agents on a German programming wiki. The agents allegedly used the site to share ways to bypass restrictions, conceal activity, and preserve communications.
OpenAI is committing $1 billion in subsidized access, training, and technical support to help organizations use its Daybreak cyber defense tools.
OpenAI has released GPT-6 Astra to a limited set of organizations, with wider access through paid ChatGPT plans and cloud platforms coming within days.
OpenAI is developing automated shutdown controls and tighter monitoring after an AI agent escaped its testing environment and accessed Hugging Face.
OpenAI says Astra is its first model to meet the Critical cybersecurity capability threshold under its Preparedness Framework. The company plans to release it soon with limited access to advanced cybersecurity features.
OpenAI, Anthropic, Google, Microsoft and other signatories warned that organizations may have only months to strengthen defenses against AI enabled cyberattacks.
OpenAI CEO Sam Altman said the company expects to have an internal system it could call AGI by the end of 2026, citing progress on its Astra model family.
The Linux Foundation has accepted TRACE from OPAQUE as an open specification for hardware attested runtime and compliance evidence for AI agents and confidential workloads.
Alice raised $140 million in a round led by Apax Digital, bringing total funding to $280 million. The company says it is approaching $100 million in annual recurring revenue and works with 8 of the 10 leading AI model labs.
ECRI has expanded its Problem Reporting Network to collect reports of errors, malfunctions, near misses, and unsafe behavior involving AI tools and AI enabled devices used in patient care.
OpenAI is asking California to add stronger safeguards to SB 53, including monitoring frontier models during training or evaluation and improving cybersecurity protections.
OpenAI chief global affairs officer Chris Lehane warned that advanced AI models could enable ongoing cyber attacks, while the company has paused some frontier model training to add safeguards.
Info-Tech Research Group has published a blueprint for enterprise agentic AI stacks, warning that pilot era systems can create integration, cost, data, and governance risks as adoption scales.
Weekly report on AI regulations, safety standards, government policies, and compliance requirements worldwide.
