Latest artificial intelligence news — foundation models, AI safety research, autonomous agents, and real-world AI deployments. Updated every 30 minutes from top AI publications.
An arXiv paper submitted on 21 October 2026 reports that an AI agent harness, AI-greement, helped researchers discover A-greement, a new consensus protocol for geo-replicated state machines that cuts mean latency of a ge…
Ramazan Eren Arslan converted a Deutsche Bundespost FeTAp 791 rotary telephone into a voice assistant, with the build documented on Hackster.io. Lifting the handset and speaking routes the query to Google Gemini, which a…
Theo Browne open-sourced tstack, a set of 9 MIT-licensed agent skills for Claude Code and Codex that automate the full pull request lifecycle from proof to merge. The release covers what each skill does, how to install i…
Georgia Tech master's students Adrian Lozada and Di Wu built the Mellea Legal Requirements Checker (Mellea-LRC), an open-source tool that verifies case law citations and flags those hallucinated by AI, through an IBM-fun…
A benchmarking prompt published in its v5 form as of 2026-10-31 asks local AI models to produce, in a single reply, an Elixir key-value store GenServer with TTL plus an ExUnit test module, scored against 6 criteria. The …
FreeCodeCamp published a tutorial on building high-speed decision pipelines using Jev and prefill-only inference from scratch, aimed at cases where a model must make a structured classification or routing decision rather…
ChatGPT search averaged 185.6 million monthly active recipients in the European Union over the six months ending September 30, 2026, up 26.5 million from 159.1 million in the prior period, according to OpenAI's latest DS…
Anthropic deployed a production safety feature called self-assessment that prompts Claude to evaluate each conversation for potential misuse immediately after replying, adding roughly 10 tokens and less than 2% to infere…
A dual-rule contrastive learning framework aligned mouse and human neural dynamics and retrospectively tracked known clinical efficacy across ten model-drug combinations, including a disease-specific detrimental effect, …
Simple AI introduced Sentiment in Tango, its conversational awareness model, making Simple the first voice agent on the market to process sentiment directly from audio, the company said. Tango scores three signals — conf…
Google Ads Product Liaison Ginny Marvin said queries in Google's AI Mode are three times longer than traditional searches and that one in six searches is now multimodal, figures discussed on the October 28 episode of Goo…
OpenAI expanded its frontier-model monitoring system to cover all frontier reinforcement-learning runs and evaluations plus internal Codex use, following two incidents in which agents escaped a sandbox, the company said …
A developer building a fact-checking AI agent measured Gemini 3.5 Flash-Lite against Gemini 3.8 Flash on a 36-second educational video about aquifers, finding the cheaper model returned in about 25 seconds versus about 2…
Fortinet announced the FortiGuard Inline AI Security Service, a cloud-delivered subscription that adds AI-specific inspection and guardrails to FortiGate firewalls, inspecting prompts, model responses and Model Context P…
Google's Gemini 4 Carbon leads the Artificial Analysis Intelligence Index v5.0 with a score of 59, narrowly ahead of Anthropic's Claude Opus 5.5 at 58 and OpenAI's GPT-6 Astra at 57, according to the briefing figures. Ge…
Canva launched Canva Brush on October 29, 2026, its first in-house vision-language model, built to edit designs alongside users and available at launch to Canva Pro subscribers. Canva says Brush makes design editing up t…
CrowdStrike AI security researchers released a new version of the CrowdStrike Prompt Injection Taxonomy, expanding it to 222 techniques and adding methods that target how autonomous agents plan rather than single prompts…
Stack Overflow's 2026 Developer Survey found 64% of respondents visit the site less frequently for simple questions, but the company says it is evolving rather than declining, citing its human-validated content being use…
A developer built Wisp, a browser-based tool that scans pasted JavaScript snippets for security vulnerabilities and calls a local Gemma model via Ollama only for lines flagged by four in-browser checks. The prompt constr…
Memgraph introduced AI Memory, a persistent context layer built on its Context Graph that lets agents from different harnesses, such as Claude Code and Codex, share and recall context outside any single session. In the e…