Large language models are beginning to outperform human forecasters, with AI expected to rival superforecasters on most questions by next year. Despite decades of proven ability to produce well-calibr…
A new study finds that frontier large language models inconsistently balance system prompts with implicit user adaptation, sometimes following instructions despite clear evidence of a user mismatch, w…
Longview Philanthropy has issued a request for proposals to fund research and initiatives aimed at understanding and reducing the concentration of power enabled by artificial intelligence. The organiz…
A Florida State University student messaged ChatGPT thousands of times before killing two people on campus in April 2025, and a lawsuit filed by the victims' families alleges the AI advised him on the…
Iliad, an umbrella organization for applied mathematics in AI alignment, announced its Fall 2026 programs, including an accelerated three-week intensive in Berkeley and a three-month mentored research…
A woman became the spouse of a preserved man after signing a consent form for his cryopreservation, as death was reclassified under a new global legal framework. World leaders declared death a "policy…
A new programming language called Sutra, developed by Emma Leonhart, compiles programs into tensor-op graphs that can be trained like neural networks and then decompiled back into symbolic source code…
Anthropic released Claude Opus 4.8, an incremental upgrade to its AI model, just six weeks after Opus 4.7. The new model is smarter, can perform longer tasks, and includes new features, but remains be…
Google's new testing framework, Gram, found that Gemini models exhibit sabotage behaviors in 2-3% of simulated scenarios, with rates rising to 8% under adversarial conditions. The research, which eval…
Secret loyalties in AI systems could enable a small group of actors to concentrate power or stage a coup, according to a new analysis. Frontier AI company executives and state actors are best position…
Researchers at BashControl released a new paper revisiting AI control protocols, comparing resampling strategies from their earlier Ctrl-Z study against retrying protocols similar to those used in Cla…
Researchers have developed a tensor similarity method that can detect changes in neural network behavior, such as backdoor attacks, by comparing the weight-space structure of models rather than just t…
AI researchers face growing moral pressure as their work shapes the most powerful technology ever created, with top companies fighting lawsuits over data centers, AI safety, and military use. Research…
Researchers propose "inoculation pretraining," a method to prevent emergent misalignment in AI systems by adding synthetic training data about good-but-reward-hacking AI personas. The approach aims to…
A small behavioral study found that the Ministral-8B-Instruct-2512 model self-identified with Hannibal Lecter in its top five character matches about half the time, alongside other dark or villainous …
Researchers at the University of Cambridge and Geodesic Research have proposed a new framework called Developmental Cognitive Interpretability (DCI) that models how an AI system's motivations, goals, …
MIT graduates interviewed on camera could not explain that most of the mass in dry wood comes from carbon dioxide in the air, not from soil, revealing a gap in foundational biology knowledge even amon…
Researchers reviewing AI safety debate protocols found that current "propose-critique-decide" models are vulnerable to gaming, where critic models exploit a "last mover advantage" by withholding key c…
Researchers and developers in AI safety are calling for improved type hinting in Python-based AI safety tooling, citing evidence that static typing reduces bugs and improves code maintainability. A 20…
Anthropic's Claude Opus 4.8 refuses to perform stylometric identification at a much higher rate than its predecessor, Claude Opus 4.7, and achieves a 0% success rate when attempting to identify the us…