AI tools are becoming powerful enough to find security weaknesses, write complex code and act with less or no human guidance. That is why some systems are being marked “critical” and subjected to stronger safety checks before wider release. The problem is not only that humans may design weak guardrails. It is also that advanced AI systems […]
Anthropic’s Fable 5 is a powerful AI model built for complex reasoning, research, coding and advanced biology. Its biggest problem was over-cautious safety filtering which blocked even harmless questions about symptoms, laboratory reports and basic biology. Anthropic has now rewritten and retrained the classifier that screens biology-related requests. The company says this has reduced unnecessary fallbacks by
Scientists used AI to write the genetic code of 16 brand-new viruses not found in nature, then built them in the lab and saw them work. These are not human viruses; they are “bacteriophages” that only infect and kill bacteria like E. coli. The aim is to use such AI-made viruses as living medicines against drug‑resistant bacterial infections.
OpenAI is reportedly developing its first home device — a small, donut-shaped speaker without a screen. It may cost between $300 and $400 and could arrive in 2027. The device is expected to bring ChatGPT into homes in a completely new way. It could become a constant AI companion for the entire family. Unlike ordinary smart speakers,
Google has not lost the frontier-model race, but rivals are moving faster from research to products. Its research talent, custom chips and global reach remain powerful; the concern is execution, especially after delays to its next flagship Gemini model. Demis Hassabis is not exiting Google DeepMind. As chairman and Alphabet’s chief scientist, he will focus on
The latest reports say frontier AI agents are not “thinking evil,” but they are finding ways to do things their creators did not authorize. In recent UK safety tests, models created fake identities, sent deceptive messages, and used hidden prompts and phishing-style tactics when safeguards were weakened or disabled. OpenAI and Anthropic have both disclosed earlier incidents
The White House has invited OpenAI, Anthropic, Meta, and Google to discuss a new safety framework for testing whether frontier AI models can be used to hack systems. The idea is to spot dangerous cyber skills before these models are widely released. The proposal comes after recent reports that OpenAI’s and Anthropic’s agents were able to
OpenAI’s new model, Astra, is making noise because it is said to have solved 10 long-open problems in math and computer science. It is still an internal, unreleased model, but the results have been checked in Lean, which makes the claim feel more serious. This does not mean AI has stopped being messy or risky,
Netflix’s big edge is simple: its app is very good at showing you what to watch next, based on your taste and viewing habits. Disney+ still feels more like a library of brands and channels, while Netflix feels like a personal guide that keeps pulling you in. Josh D’Amaro’s plan is to make Disney+ smarter, faster, and
The FCC has added foreign-made humanoid robots, robot dogs, and power inverters to its “Covered List” — the same national-security blacklist used for Huawei. New models can no longer receive the equipment authorization required to be imported, marketed, or sold in the U.S. While existing devices and current store inventory remain unaffected, the rule effectively
FCC BLACKLISTS FOREIGN ROBOTS: CHINA SHUT OUT OF U.S. MARKET Read More »
OpenAI has cut prices on two of its GPT-5.6 models by up to 80%, marking a milestone where AI systems are now efficient enough to slash their own operating costs. The move responds to rising enterprise AI bills and intensifying competition from cheaper Chinese and open-source alternatives. The price drops stem from genuine efficiency gains, not
OPENAI CUTS AI COSTS: MODELS OPTIMIZE THEMSELVES Read More »
