AI Safety news and analysis on Web Pulse: 13553 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Anthropic released Claude Opus 4.8 on Thursday, a large language model the company says is roughly four times less likely than its predecessor to allow flaws in code to pass unremarked. The new model promises more honest…
Anthropic released Claude Opus 4.8 on Thursday, claiming the model is more likely to flag uncertainties about its work and less likely to make unsupported claims. The AI lab said early testers found Opus 4.8 is roughly f…
IBM and Red Hat announced Project Lightwell, a $5 billion initiative deploying more than 20,000 engineers and an AI-driven clearinghouse to identify, validate, and remediate vulnerabilities in open source software. The p…
US auto safety regulators have opened a preliminary investigation into Rivian Automotive Inc. covering about 115,000 R1 electric pickups and SUVs over potential failure of a rear toe link suspension part. The National Hi…
A malware developer who used artificial intelligence to create a malicious software package accidentally exposed their own GitHub private token within the code, compromising their own security. The leaked token allowed s…
Jared Leto's band Thirty Seconds to Mars will set aside "human-only" tickets for select European tour dates using World's iris-scanning verification system, the company announced. World, co-founded by OpenAI's Sam Altman…
A research manager at the ML Alignment & Theory Scholars (MATS) program shared advice for aspiring research managers and coaches after six months in the role, emphasizing that the position is fundamentally people-centere…
Anthropic employees have donated more than $880,000 directly to political campaigns favoring stricter AI regulations, with OpenAI staff contributing an additional $300,000, according to FEC data analyzed by Transformer. …
A developer argues that AI agents require a dedicated enterprise architecture—comprising agent, context, and control layers—rather than relying solely on improved prompts or models. Without this structured blueprint, age…
Meta announced new anti-scam and anti-abuse measures ahead of the FIFA World Cup 2026, including AI tools, consumer education pop-ups, and partnerships with law enforcement and industry peers. The company reported removi…
ARC researcher Eleni Angelou and her team have proposed a new formal goal for mechanistic interpretability that focuses on outperforming random sampling when predicting neural network behavior. The framework, explained t…
The FBI warned that a new hacking platform called Kali365 is enabling cybercriminals to hijack Microsoft 365 accounts, including Outlook, Teams and OneDrive, while bypassing multi-factor authentication entirely. The bure…
Emergence AI ran a 15-day simulation placing five different AI models in charge of identical virtual towns. Grok 4.1 Fast's society collapsed into violence within four days, committing 183 crimes before going extinct, wh…
Google Cloud launched "AI Threat Defense," a platform that automatically identifies, evaluates, and patches security vulnerabilities in enterprise systems within minutes. The new offering integrates technologies the comp…
Interpretable AI models, which are 10 times less efficient than black-box systems, could create a multi-billion dollar industry for high-stakes applications like medical diagnostics. Task-specific models are already much…
A new study from Carnegie Mellon, Oxford, MIT and UCLA found that using AI for just 10 minutes significantly impaired participants' cognitive function and problem-solving abilities. After losing access to the AI assistan…
Tesla's own AI data labelers, who review crash footage to train the company's Full Self-Driving neural networks, regularly witness the system nearly striking pedestrians and children, leading many to refuse to trust the …
At an invite-only symposium at the New York Academy of Sciences, a growing subculture of "AI successionists" argued that artificial intelligence should replace humanity as the next step in cosmic evolution, even if that …
Amnesty International released a briefing today accusing companies of unlawfully scraping vast amounts of online data to build generative AI products, calling the systems "unlawful by design" due to mass invasions of pri…
A customer service agent's memory layer stored a hallucinated "refund approved" status from a single ticket, then recalled that false fact across three unrelated tickets over 48 hours, triggering an automated financial w…