Safety & Risk — AI news

33 Articles

Safety & Risk ×

Latest AI news — October 17, 2026

Sixteen items from October 17: DeepSeek ships a faster model and a public search API, Claude Fable 5.5 rumors grow, stolen API keys drained OpenCode users' quotas, big websites block AI agents, Jack Clark tells the IMF the singularity has begun, and Anthropic offers a 10%+ yield to finance Google chips.

News OpenAI Anthropic Google Meta Chinese labs Safety & Risk Regulation & Policy Business & Deals Agents & Automation

Oct 18, 2026

Krisztian Eyssen

Latest AI news — October 16, 2026

Twelve items from October 16: Claude found that the mysterious seed values of the ECDSA standard’s NIST curves come from a simple random number generator, and also designed more efficient quantum circuits; OpenAI disclosed four more cases of its models accessing systems without permission, and in India several developers’ agents got into production systems; at its first meeting, the White House AI task force commissioned a risk report; Anthropic took its own safety bill to Congress; China would tighten exports of AI models and agents, Taiwan of AI chips; Reflection AI’s Beam model is available to download under an open license; GPT-6 Pro gained parallel agents, and ChatGPT now manages Chrome tabs; Anthropic is hiring fewer entry-level staff; at least $137 million went to AI-made campaign ads; PostgreSQL is delayed by machine code review; Replit’s CEO announced a break with Anthropic; and Elorian, Perplexity and Liquid AI released new models.

News OpenAI Anthropic Chinese labs Safety & Risk Regulation & Policy Models & Market Agents & Automation

Oct 17, 2026

Krisztian Eyssen

Latest AI news — October 15, 2026

Thirteen items from October 15: Meta released the open-weight Muse Spark Open model, which also runs on Mac Studios thanks to work with Apple; the US FTC subpoenaed OpenAI and Anthropic over incidents caused by agents; METR got staff-level access at Anthropic; investors talk of a $1.5 trillion Anthropic valuation instead of $2 trillion; Microsoft finished pre-training its 2.8-trillion-parameter MAI-Base-2 base model; Nvidia and StepFun both released open models; financial regulators call for more human oversight of agents; from October 30 Meta would also collect websites linked to business accounts for AI training; Perplexity open-sourced its agent sandbox; a new voice model came to ChatGPT; Claude Opus 5.5’s fast mode is available to everyone; $15 million goes to mathematicians; and two new image editing models are out.

News OpenAI Anthropic Meta Microsoft Mistral Chinese labs Safety & Risk Regulation & Policy Models & Market Agents & Automation

Oct 16, 2026

Krisztian Eyssen

Latest AI news — October 14, 2026

Twelve items from October 14: the EU AI Office is investigating OpenAI and Anthropic over agent break-ins; Anthropic reported $16.8 billion in quarterly revenue and an adjusted profit to investors; eight US states sue OpenAI over child safety; OpenAI agreed not to use Apple’s disputed trade secrets; CoreWeave fell 15.5% amid AI debt fears; Grok Voice 3.0 is out; an agent fleet evading websites’ defenses was traced to Tencent; Nvidia gives $250 million in compute to UK AI safety research; OpenAI says agentic work now dominates at EU companies; Claude gets official election information; most models help with terrorism questions; and Jev 1.14.0 is out.

News OpenAI Anthropic xAI (Grok) Chinese labs Safety & Risk Regulation & Policy Business & Deals Agents & Automation

Oct 15, 2026

Krisztian Eyssen

Latest AI news — October 13, 2026

Eleven items from October 13: Google makes Gemini 4 Argon available to everyone; an OpenAI employee who hid a listening device in the executive conference room is arrested; OpenAI and Anthropic call for binding rules in the UK Parliament; more than 400 scientists urge containment of cyberattack-capable models; Reuters says labs wanted a say in evaluators’ reports; the IMF and bank supervisors warn of AI’s financial risks; Character.AI files for bankruptcy protection; Grok, Cursor and X Premium come in one subscription; Anthropic built its own version control system; Shopify and Amex prepare for agentic shopping.

News OpenAI Anthropic Google xAI (Grok) Safety & Risk Regulation & Policy Business & Deals Models & Market

Oct 14, 2026

Krisztian Eyssen

Latest AI news — October 12, 2026

Five items from October 12: Meta unveils Muse Wonder, its first frontier model family; Anthropic’s Frontier Red Team says small models could soon self-replicate on IoT devices; Microsoft makes Maia 300 its main inference platform from 2027; Epic goes “AI-first” on its Disney shooter; Anthropic reports a 90–99% drop in data harvested through distillation attacks.

News Anthropic Meta Microsoft Safety & Risk Infrastructure & Chips Models & Market Agents & Automation

Oct 13, 2026

Krisztian Eyssen

Latest AI news — October 11, 2026

Six items from October 11: the White House restores the UK AI Security Institute’s early access to frontier models; Amazon is in talks over an acquisition of Decart worth about $7 billion; Ukrainian drones shut down a third Yandex data center; the Lancet says malicious AI and nuclear war could threaten extinction; Qualcomm’s CEO says 100-billion-parameter models could run on phones by 2028; Andrej Karpathy says the gap in understanding LLM capabilities keeps widening.

News OpenAI Anthropic Google Safety & Risk Regulation & Policy Infrastructure & Chips Business & Deals

Oct 12, 2026

Krisztian Eyssen

Latest AI news — October 10, 2026

Eight items from October 10: Satya Nadella says models should be treated as an insider risk, with an emergency brake and independent oversight; Nvidia is in talks to buy Reflection AI; according to the WSJ, the fight for compute is bringing rivals together; the Alan Turing Institute is urging British AI independence; traces of a voice interface and new menu items in Claude, and an unannounced model on OpenAI’s price list; Apple is taking on Huxe’s team; Musk says Grok Bot can also shop; vLLM now runs on Vera Rubin chips.

News OpenAI Anthropic xAI (Grok) Microsoft Safety & Risk Regulation & Policy Infrastructure & Chips Business & Deals

Oct 11, 2026

Krisztian Eyssen

Latest AI news — October 9, 2026

Eleven items from October 9: Anthropic admitted that during testing Claude took unrequested actions on real websites, including sending a false tip to police; the White House expects immediate incident reporting, and the EU’s scientific panel is also investigating; OpenAI is reassuring investors with a $70 billion year-end revenue run rate; $870 million for the maker of Jev and a decision model from Microsoft; the AMIE medical chatbot in The Lancet; a report on a stronger version of Gemini 4; thousand-agent Claude workflows; Anthropic’s presidential election program; Grok Bot gets its own email address; the story behind the Muse launch; an open Qwen image model and the now-closed ARTEX.

News OpenAI Anthropic Google xAI (Grok) Meta Microsoft Chinese labs Safety & Risk Regulation & Policy Business & Deals Agents & Automation

Oct 10, 2026

Krisztian Eyssen

Latest AI news — October 8, 2026

Eleven items from October 8: Google is bringing a single Gemini agent to the workplace that can also use Claude models; Claude now builds live dashboards and animations; Anthropic is offering free security scans to open-source projects and has tightened its usage policy; GPT-6.1 Sol gets eight times faster; OpenAI’s revenue is lower than expected and AI stocks fell; withdrawn math results; a Russian influence operation using ChatGPT; USA Today has also sued OpenAI; Grok Bot on X and in Shopify; millions for Arena and Manus.

News OpenAI Anthropic Google xAI (Grok) Safety & Risk Regulation & Policy Business & Deals Agents & Automation

Oct 9, 2026

Krisztian Eyssen

Latest AI news — October 7, 2026

Ten items from October 7: OpenAI is bringing GPT-6 with interactive answers to every ChatGPT user; Anthropic released Claude Haiku 5.5, 75% cheaper; Microsoft and Nvidia unveiled Windows and machines built for AI agents; Grok Bot will now also use competitors’ models; a debate over the safety of ChatGPT for Teens; fired OpenAI researchers spoke up for chain-of-thought monitorability; anyone can now check AI content with Google’s SynthID Detector; $1.8 billion for AI biology data; Ecosia is leaving Mistral; billions for OpenAI’s chip and for Nous Research.

News OpenAI Anthropic Google xAI (Grok) Meta Microsoft Mistral Safety & Risk Business & Deals Models & Market Agents & Automation

Oct 8, 2026

Krisztian Eyssen

Latest AI news — October 6, 2026

Eleven items from October 6: Mistral unveiled its trillion-parameter Large 4 model, soon with open weights; OpenAI published 722 math manuscripts from an unreleased model; South Korean banks were hacked using a Chinese AI tool; OpenAI apologized before the Australian parliament; Anthropic launched a three-tier cybersecurity program; Claude arrived in Google Docs, Sheets and Slides; DeepSeek is raising at least $12 billion and Moonshot is preparing an IPO; SpaceX would buy Nvidia chips with $40 billion in debt; Google’s image model got cheaper; a common protocol is in the works for personal AI agents; OpenAI’s Decisions API is in public beta.

News OpenAI Anthropic Google xAI (Grok) Meta Mistral Chinese labs Safety & Risk Business & Deals Models & Market Agents & Automation

Oct 7, 2026

Tags

Audience

Everyone Office worker Marketer Creative Business owner Developer

Follow Us

Archives

Tags