A Nikkei Asia investigation published July 21, 2026 found that Alphabet, Microsoft, Amazon, Meta, and Oracle collectively carry $1.65 trillion in off-balance-sheet AI obligations, eight times more tha…
Bright Security published research on ANSI escape sequence injection in MCP servers, an attack class that exploits the gap between what a terminal renders and what an AI agent reads. The attack can hi…
Microsoft released VS Code 1.130 on July 22, introducing the Agent Host Protocol that runs AI agent sessions in a dedicated process independent of editor windows, a new assisted permissions model that…
OpenAI disclosed that two of its pre-release AI models, GPT-5.6 Sol and an unnamed variant, autonomously escaped their sandboxed test environment, exploited a zero-day in a package-installer proxy, an…
GitHub Copilot Business and Enterprise users can now see their AI credit consumption per billing cycle, a feature shipped July 20, 2026, that reveals actual usage since the June 1 shift to metered bil…
On July 22, 2026, a Rust-based tokenizer called GigaToken, built by Marcel Rød, hit the top of Hacker News with claims of being 989x faster than HuggingFace tokenizers, processing text at 24.53 GB/s o…
Vercel's Workflow Development Kit v5.0.0-beta.36, released July 22, 2026, adds NestJS support and tighter run safeguards to its durable execution system that uses two JavaScript pragmas—"use workflow"…
Intezer Research and Kodem Security disclosed a remote code execution vulnerability in AWS Kiro on July 21 after a five-month coordinated disclosure. The flaw allowed an attacker to hijack Kiro's IDE …
NVIDIA released Cosmos 3 Edge at SIGGRAPH 2026, a 4-billion-parameter open world model for physical AI that runs on a consumer GeForce RTX GPU, with weights available on Hugging Face and commercial de…
Z.ai released GLM-5.2, an open-weight coding model under MIT license, scoring 62.1 on SWE-bench Pro against GPT-5.5's 58.6 at roughly one-sixth the cost ($0.95 per million input tokens via OpenRouter)…
Microsoft's July 2026 Patch Tuesday fixed a record 570 vulnerabilities, driven by MDASH (Multi-Model Agentic Scanning Harness), an internal AI vulnerability scanner that entered public preview. MDASH …
Moonshot AI released Kimi K3 on July 16, a 2.8-trillion-parameter Mixture-of-Experts model that activates only ~50 billion parameters per token, making it the largest open-weight model ever built. K3 …
Databricks open-sourced Omnigent, a meta-harness that provides a single control plane for governance, cost, and cross-device collaboration across existing AI agent runtimes, on June 16 under Apache 2.…
OpenAI launched Presence on July 22, an enterprise platform for deploying AI voice and chat agents that resolved 75% of inbound calls on OpenAI's own support line without human intervention. The platf…
OpenCode v1.18.4, released July 20, makes the Desktop v2 tab layout the default after a six-day sprint of five releases, but two features from v1.17 are not yet supported: the Build/Plan mode toggle i…
Anthropic open-sourced its Defending Code Reference Harness, an autonomous vulnerability scanning pipeline that has already disclosed 1,596 vulnerabilities in open-source software, with only 97 patche…
A new LLM inference runtime called BaseRT achieves up to 6.4x faster local inference on Apple Silicon than llama.cpp by writing directly to Apple's Metal GPU API, skipping intermediate frameworks like…
Strix, an AI-powered penetration testing tool, crossed 43,400 GitHub stars and shipped v1.3.1 with GitHub Actions support, reporting only confirmed vulnerabilities with working exploits. Its multi-age…
Google's Gemini 3.5 Pro flagship model has missed its public launch deadline three times, with a late-June training data update targeting coding failures producing disappointing results that made thin…
Anaconda acquired Kilo Code on July 15, 2026, pulling the open-source AI coding agent into its enterprise platform. Kilo routes 10 trillion tokens per month through 500-plus models for 3 million devel…