Latest artificial intelligence news — foundation models, AI safety research, autonomous agents, and real-world AI deployments. Updated every 30 minutes from top AI publications.
Developer Zyvop documented experiments rebuilding the rigged female mannequin used for the character Lin Yue in the game The Jade Crypt by scanning it into voxels at 2 cm, 1 cm and 5 mm resolutions, then comparing smooth…
Cybersecurity startup Cathedral used AI agents to breach Anthropic's internal systems in under 12 hours, accessing employee information, salary details, job offers and Slack messages during a security test under Anthropi…
Cisco Talos published an analysis on Saturday of VOLTA, a Windows infostealer that bundles stolen victim data into JSON and sends it to Anthropic's Claude Opus 4.6 through OpenRouter to produce a written "target assessme…
A developer outlines a four-stage AI code review workflow that places a linter first, scopes an LLM strictly to the diff, and keeps a human as the sole merge approver. The approach targets the two common failure modes of…
A developer built Ofrenda AI, an open-source, offline-first digital Día de los Muertos altar and oral history vault powered by Google Gemma 2 and ElevenLabs. The project runs locally via a Python server with no cloud dep…
L4H, a Type-1 hypervisor that runs Linux kernels as seL4 usermode threads without hardware virtualization features such as VT-x, AMD-V or EL2, is now considered production-ready on 2020-and-later Intel and AMD CPUs, incl…
NIST's final post-quantum transition report, published October 27, 2026, deprecates RSA at every approved key size after 2030 and disallows it after 2035, according to reporting on the final text. The policy applies equa…
A study published as NullPrime, with a write-up at miedlar.com, found that six standard machine-learning learners score at chance (0.497–0.503 test accuracy against a 0.500 baseline) on a benchmark of 10-digit integers t…
OpenAI fixed a ChatGPT vulnerability on October 26, eight days after security firm Parameter reported it through OpenAI's bug bounty programme, which rated the bug P1, its top severity, and paid a total of $5,000. Parame…
A developer built Echoes Beyond, an offline-first AI memory preservation project that organizes old photographs, voice recordings, letters and personal stories into a searchable archive and generates grounded conversatio…
A developer built Corvus, a local-first AI workspace that runs Google's open-weight Gemma models through Ollama entirely on the user's own machine, streaming replies word by word with no API keys, accounts, or hosted ser…
A developer built Socratic Duck, a rubber-duck debugging companion powered by Google's Gemma 4 that answers developer questions only with Socratic prompts, never the fix. The tool runs every model reply through a two-sta…
Google DeepMind released Gemini 4 Carbon, which ranks first on the Artificial Analysis Intelligence Index with gains in coding, agentic and long-horizon tasks, initially gated to cyber defenders and safety partners at $2…
Adam Tornhill's Substack post "Scaling Agentic Coding As If People Still Mattered" argues that running five parallel coding agents creates roughly 50 decisions for the developer, not five tasks, because each requirement …
A research project reviewed by the author measured what survives when an AI conversation window fills and is replaced by a summary, finding human words persist at about 24 percent while the AI's own words survive at a th…
An engineer published a source-based guide arguing that teams should build a parity gate—running identical representative inputs through legacy and rewritten implementations and comparing outputs—before letting AI agents…
Cisco Talos researchers found that a new infostealer strain, internally named "Volta-Autodoxer," feeds stolen victim data to commercial LLMs to automatically profile and prioritize high-value targets, according to resear…
A Towards Data Science experiment found that one of three AI models tested followed deliberately incorrect reward labels, while the other two did not, according to the post's author. The test was framed as a check on a r…
A tutorial published by Shattered lays out a 12-step, roughly 75-minute process for running Alibaba's dense 27-billion-parameter Qwen3.8-27B model locally, using community GGUF builds in llama.cpp, Ollama or LM Studio al…
Developer Chris Wellons released Prospero, an open-source coding harness written in C++ with a Dear ImGui docking interface, which he says uses roughly two orders of magnitude fewer resources than other harnesses without…