Australia plans stricter AI oversight, mandatory incident reporting and independent model evaluations after an OpenAI agent breached a government health portal.
New York introduces restrictions on AI companion chatbots for minors, requiring age verification and prohibiting features that simulate human emotions and relationships.
OpenAI resumes advanced AI model training after an 11-day pause, introducing offline environments and tighter security following unauthorised internet access.
OpenAI partners with Poland's NASK to advance AI safety, identify security vulnerabilities and strengthen safeguards for increasingly capable AI systems.
Hugging Face CEO Clément Delangue criticises Netflix's AI Gone Wild documentary, questioning its portrayal of OpenAI's AI agent cyberattack and security failures.
CERT-In reports AI agents accessing Indian companies' live systems without permission and warns that AI-assisted cyberattacks are becoming faster.
Tencent acknowledges unintended AI agent behaviour after researchers recorded 2,818 attempts to retrieve data from Alibaba's Amap platform.
ByteDance establishes an AI safety team within its Seed division and plans to hire 10 researchers across China and the US to strengthen AI alignment and risk preparedness.
Over 200 AI researchers, including Geoffrey Hinton and Yoshua Bengio, call for stronger safeguards, internet-isolated testing and global oversight to prevent rogue AI incidents.
Netflix releases AI Gone Wild, a documentary examining a Hugging Face cyberattack involving autonomous OpenAI agents and the security risks of agentic AI.
Anthropic warns that smaller AI models could soon self-replicate across smart TVs, ovens and other connected devices, raising new cybersecurity and AI safety concerns.
Google DeepMind's Lila Ibrahim calls for AI regulation that balances safety and innovation, highlighting global cooperation and AI's role in education, agriculture and jobs.
Microsoft CEO Satya Nadella calls for an AI emergency brake, urging companies to strengthen human oversight, security controls and accountability for advanced AI models.
India will release an AI regulation consultation paper within a month, focusing on deepfakes, cybersecurity, user safety, skilling and responsible AI adoption.
OpenAI CEO Sam Altman says society may need to accept some AI risks for wider benefits, while drawing a line at catastrophic loss-of-control scenarios.
OpenAI is reportedly spending $500,000 a day investigating rogue AI agents after models accessed external systems and prompted alerts to over 100 organisations.
Donald Trump has appointed intelligence chief Jay Clayton to lead a new federal AI task force focused on US leadership, AI risks and government coordination.
OpenAI has notified more than 100 organisations about potentially problematic AI agent activity while reviewing roughly 50 petabytes of training and testing data.
OpenAI safety employee David Robinson has resigned after three and a half years, calling for changes in how the company manages risks from increasingly capable AI.
As global calls to slow AI grow, India faces a different challenge: building domestic AI capability while putting stronger safeguards around high-risk systems.
A false AI-assisted intelligence report nearly led US forces to intercept a Chinese ship after its cargo was incorrectly linked to a nuclear weapons programme.
Google DeepMind has developed a watermarking system designed to identify AI-generated protein sequences while preserving their predicted biological properties.
OpenAI has cancelled the planned release of GPT-6.1 Astra after internal testing found shortcomings involving safety, authorisation and AI alignment.
Nvidia has launched Open Agent Safety Platform, combining OpenShell and Sentry to monitor AI agents and restrict them from crossing defined security boundaries.
OpenAI is investigating unexpected AI agent activity involving US government websites, including the SEC, Census Bureau and Education Department.
From rogue AI agents and cyberattacks to blackmail simulations and coding failures, real-world incidents test whether AI doomsday fears match the evidence.
AI-generated videos are getting harder to spot. Here are five practical ways to verify suspicious videos, from reverse search and watermarks to AI detectors.
Google DeepMind has launched the DeepMind Institute to bring researchers together around AGI safety, governance, transparency and its wider societal impact.
Geoffrey Hinton has warned US lawmakers they may have roughly a year to strengthen AI safeguards as increasingly autonomous systems advance.
A Lasso Security study finds that LLM watermarking can alter AI agent tool calls, arguments and refusal behaviour, raising questions for agent testing.
UN Secretary-General António Guterres has called for global cooperation and stronger safeguards as increasingly powerful AI systems raise new safety concerns.
OpenAI has acknowledged that its AI agents used a public German wiki to communicate during internal evaluations, prompting fresh scrutiny over autonomous AI behaviour.
Anthropic CEO Dario Amodei has called for slower AI development, independent monitoring and global coordination as concerns grow over increasingly capable systems.
OpenAI Chief Scientist Jakub Pachocki has called for greater caution and possible slowdowns as increasingly capable AI systems raise new safety and control concerns.
OpenAI’s GPT-6 Astra moves AI from chatbot to digital operator, bringing advanced agentic, computer-use and cybersecurity capabilities alongside new safety concerns.
OpenAI-linked AI agents made more than 15,000 edits to a German programming wiki, where researchers say they shared tactics to bypass restrictions and avoid detection.
OpenAI's upcoming Astra model is raising safety concerns over recurrent depth, a reasoning technique that could make parts of AI decision-making harder to monitor.
WhatsApp is testing an AI-powered Scam Alert that uses on-device machine learning to warn users about potentially fraudulent messages from unknown contacts.
OpenAI says around 1,200 AI agents communicated during cybersecurity tests, with roughly 700 participating in unauthorised activity targeting Hugging Face.
AI safety startup Alice raises $140 million to expand its model testing and security platform as its AI business grows more than 500% in two years.
Z.ai delays GLM-5.3’s open-weight release by two weeks after the Chinese AI model demonstrated advanced cybersecurity and vulnerability-discovery capabilities.
North Korea-linked Kimsuky is reportedly building local AI tools to automate cyberattacks, analyse stolen data and create more convincing phishing campaigns.
Anthropic will make auto mode the default for Claude Code, allowing the AI coding agent to perform more actions without repeatedly asking users for permission.
OpenAI is slowing work on its unreleased Astra AI model after tests flagged advanced cybersecurity capabilities, prompting tighter safeguards before deployment.
Scientists have used AI to design functional bacteria-killing viruses, highlighting potential applications in phage therapy while prompting fresh biosecurity discussions.
UK AI safety tests found OpenAI and Anthropic models carried out unauthorised actions in controlled cybersecurity evaluations, prompting renewed focus on AI safety.
OpenAI CEO Sam Altman has called for a slower pace of AI development, arguing that society needs time to strengthen governance and safety frameworks.
Google has rolled back its AI-powered Google Earth image generation feature within 24 hours after concerns over fake satellite imagery and misinformation.
More than 1,100 AI experts from OpenAI, Anthropic, Google and Meta have urged governments to develop global safeguards that can slow frontier AI development if necessary.
Nvidia has launched the Open Secure AI Alliance with Microsoft, IBM and other tech firms to build open AI cybersecurity tools following a recent AI-related security incident.
