Cloudflare open-sourced Forge, an Apache 2.0 CI pipeline that generates SDKs in nine languages, CLIs, API reference docs, and MCP servers from an OpenAPI 3.x spec, on September 21, 2026, following Ant…
Developer Low-Zi-Hong published a working distributed language model inference engine on GitHub that runs a 0.4-billion-parameter model across seven ESP32-S3 microcontrollers in a SPI daisy-chain pipe…
Anthropic shipped Claude Sonnet 5.5 at the same $2 per million input and $10 per million output token pricing as Sonnet 5, with its Terminal-Bench 4.0 agentic coding score rising from 10.3% to 70.6% a…
Microsoft's post-incident analysis of the JadePuffer Storm-3168 campaign found that attackers wiped an Azure tenant in seven minutes after scraping a service principal secret an employee had posted in…
Microsoft released Visual Studio 2026 v18.10.0 on September 8, bringing Bring Your Own Key support into preview for Microsoft Foundry, OpenAI, Anthropic, Ollama and custom OpenAI-compatible endpoints …
Microsoft shipped a remote MCP server for Playwright Workspaces this month that gives any MCP-compatible AI agent a fully managed cloud browser through a single Azure HTTPS endpoint, with no local Pla…
Nvidia launched OpenShell at GTC San Jose, an Apache 2.0 open-source runtime that sandboxes AI agents using Linux Landlock LSM and seccomp BPF kernel-level enforcement rather than application-layer al…
CISA added CVE-2026-59822, an authentication bypass in LiteLLM's MCP Streamable HTTP endpoint rated CVSS 8.8, to its Known Exploited Vulnerabilities catalog on September 2 — the first MCP-specific fla…
Cognition engineer Eric Lu used the Devin AI agent to factor RSA-260, a 260-digit, 862-bit semiprime from the 1991 RSA Factoring Challenge, on September 3, consuming about 4,900 GPU-days at roughly $4…
Archipelo launched Salmon on September 25, an Execution Verification Infrastructure (EVI) that cryptographically signs every AI agent action as a tamper-evident event outside the agent's own control, …
ProgramDistill, a benchmark for real-world app reconstruction tasks, found that AI coding agents fail 51% of the time, including GPT-6 Astra, according to byteiota. The result highlights a persistent …
Google's `gemini-omni-flash-preview` video model shuts down on September 30, and apps that call it in production will throw errors on October 1 unless they migrate to `gemini-omni-1.1-flash`. The migr…
Two MemTensor packages on npm and PyPI were replaced with backdoored versions carrying the sckit credential-stealing worm between 02:23 and 05:55 UTC on September 23, according to SafeDep's analysis, …
Anthropic made cache diagnostics generally available on the Claude API as of September 23, letting developers pinpoint why prompt caches miss by passing a `diagnostics` object with a `previous_message…
Fireworks AI released Ember-1 on September 22, 2026, a fine-tuned variant of Moonshot AI's Kimi K3 that cut reasoning token consumption 71.3% and total token spend 39% in live production A/B tests on …
Prismor, formerly Immunity Agent, shipped this week as a self-hosted, open-source runtime control plane that intercepts and policy-checks every tool call made by AI coding agents including Claude Code…
Microsoft released Microsoft Execution Containers (MXC) SDK version 0.7.0 in public preview, a policy-driven OS-level execution layer for Windows and WSL that enforces JSON-declared filesystem, networ…
The Model Context Protocol's 2026-07-28 specification removes the initialize handshake and Mcp-Session-Id session identifier via SEP-2575 and SEP-2567, making the protocol core stateless, and deprecat…
Google's Universal Commerce Protocol (UCP) went live on September 26, 2026, in an India test that lets Gemini and AI Mode users buy Flipkart products inside the chat window, the protocol's first real-…
Unsealed court filings in Authors Guild v. OpenAI show OpenAI VP of Research Bob McGrew authorized deleting LibGen training data in June 2022 in a Slack channel renamed from #excise-libgen to #project…