AI Safety news and analysis on Web Pulse: 8068 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A developer built sourcecheck, an open-source tool that verifies AI-generated citations against registries like OpenAlex and Crossref, and found that top models including GPT-5.5 and Claude Sonnet 5 cite retracted papers…
Pope Leo XIV and Cambridge astrophysicist Hiranya Peiris have independently warned that artificial intelligence poses existential risks to humanity, with Peiris describing AI systems as laying tracks ahead of a speeding …
CyberProof launched the Agentic MXDR Service, integrating AI agents with human expertise to automate threat intelligence, hunting, and detection engineering. The service uses CyberProof's Reveal360 to deliver quantifiabl…
Two 15-year-olds were arrested in San Mateo after a Waymo autonomous vehicle they were riding in alerted police to their misbehavior, including drinking alcohol and shooting water beads from a toy gun. The car stopped in…
Google's Threat Intelligence Group reported on May 11, 2026, the first documented case of criminals using an AI model to discover a zero-day vulnerability and develop a working exploit, targeting a two-factor authenticat…
Picus Security launched the Picus Autonomous Exposure Validation Platform to address the rapid weaponization of CVEs by adversaries, enabling organizations to validate real-world exploitability of vulnerabilities against…
Bespoke Labs, a Mountain View startup building simulated environments to train and test AI agents, raised $40 million in a Series A and seed round led by Wing VC and 8VC, with backing from executives at Anthropic, OpenAI…
Anthropic's $65 billion Series H round, led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, accounted for nearly half of Q2 2026 venture funding, highlighting extreme concentration in AI investment. The …
An AI must receive information about its environment and have ways to affect it to solve real-world tasks. The concept of 'actual entanglement' measures how much information an AI has about its environment, while 'minimu…
A developer discovered that standard agent logging schemas hide failures by logging what the model says rather than what actually happened. After 12 weeks of iteration, they developed a four-field schema—expected_outcome…
A developer outlines a secure method for managing API keys and secrets on macOS using Keychain for machine-level secrets, the 1Password CLI for project-specific secrets, and direnv for automatic loading, after a close ca…
Tech workers are splitting into two camps: those amplified by AI and those shaken by it, according to a 2026 survey. Burnout surged 11 points to 55.7%, career optimism fell to 48.7%, and 53% would not recommend their fie…
Elon Musk's proposal for 'Universal High Income' driven by AI and robotics was met with a stark warning from investor Michael Burry, who predicted revolution before such a system could be implemented. The exchange highli…
A UPI Voices essay published on July 7, 2026, argues that humanism, transhumanism and posthumanism are reshaping debates over what counts as the human person, offering conceptual framing for AI ethics and human-centered …
Varonis Threat Labs discovered a critical vulnerability in Google Cloud Platform's Dialogflow CX service, allowing attackers to hijack AI conversations, steal data, and launch further attacks. The flaw, dubbed 'Rogue Age…
Harry Shearer, known for 'The Simpsons' and 'Spinal Tap', and writer Tom Leopold are premiering their musical 'Here Comes J. Edgar' at London's King's Head Theatre on July 10, after three decades of development. The show…
A DigiCert survey of 1,001 IT and security leaders found 78% of organizations experienced AI-related security incidents or vulnerabilities, with 27.7% reporting one incident and 21.9% multiple incidents. The report highl…
Companies that track AI usage risk turning metrics into weapons, as seen with token leaderboards that spawned 'tokenmaxxing' behavior. The problem is not measurement itself but the culture that uses it, where metrics bec…
Commvault launched Commvault Minutes to Recovery, a cyber resilience simulation that lets participants use AI tools to launch attacks and then defend against them. The window between vulnerability discovery and exploitat…
OpenAI restricted its GPT-5.6 Sol AI model to government-approved partners after a request from the Trump administration citing safety concerns, marking the first time a U.S. lab has launched a frontier model under such …