https://bvdnet.nl/en/blog Background on AI models, agents and security, with sources. en Tue, 03 Nov 2026 19:41:53 GMT https://bvdnet.nl/en/blog/what-left-pad-teaches-about-supply-chain-attacks https://bvdnet.nl/en/blog/what-left-pad-teaches-about-supply-chain-attacks Mon, 03 Aug 2026 00:00:00 GMT In 2016 one maintainer unpublished eleven lines and broke builds worldwide. The axios and node-ipc attacks reuse that same mechanism. AI Intel Pipeline Safety & Ethics https://bvdnet.nl/en/blog/what-cnc-machining-teaches-about-agentic-workflows https://bvdnet.nl/en/blog/what-cnc-machining-teaches-about-agentic-workflows Thu, 30 Jul 2026 00:00:00 GMT A CNC part goes from stock to finished through a fixed sequence: fixture, rough, finish, measure. Agentic coding workflows follow the same order. AI Intel Pipeline Agentic AI https://bvdnet.nl/en/blog/anthropic-project-glasswing-ai-vulnerability-discovery https://bvdnet.nl/en/blog/anthropic-project-glasswing-ai-vulnerability-discovery Wed, 27 May 2026 00:00:00 GMT Anthropic and about 50 partners found over ten thousand serious flaws with Claude Mythos Preview in a month, and fixing them is now the bottleneck. AI Intel Pipeline https://bvdnet.nl/en/blog/google-io-2026-gemini-omni-information-agents https://bvdnet.nl/en/blog/google-io-2026-gemini-omni-information-agents Wed, 27 May 2026 00:00:00 GMT At I/O 2026 Google showed Gemini Omni, which makes video from any input, and agents that keep working in the background in Gemini and Search. AI Intel Pipeline Agentic AI https://bvdnet.nl/en/blog/self-evolving-ai-models-minimax-m2-recursive-improvement https://bvdnet.nl/en/blog/self-evolving-ai-models-minimax-m2-recursive-improvement Wed, 27 May 2026 00:00:00 GMT MiniMax says M2.7 takes an early step toward self-evolution: it debugs its training runs and edits its own agent scaffold, while people set the goals. AI Intel Pipeline Models & Architecture https://bvdnet.nl/en/blog/why-real-world-ai-agent-performance-lags-behind-hype https://bvdnet.nl/en/blog/why-real-world-ai-agent-performance-lags-behind-hype Wed, 27 May 2026 00:00:00 GMT On JobBench and ITBench-AA, two May 2026 benchmarks built from real work, the best AI agent scored below 50%, while older work tests were near saturation. AI Intel Pipeline Agentic AI https://bvdnet.nl/en/blog/how-do-multi-agent-systems-learn-from-failures-without-retraining https://bvdnet.nl/en/blog/how-do-multi-agent-systems-learn-from-failures-without-retraining Tue, 19 May 2026 00:00:00 GMT In EVOCHAMBER (May 2026), AI agents pass lessons from strong to weak after a failure and form specialists, without changing the model's weights. AI Intel Pipeline Agentic AI https://bvdnet.nl/en/blog/which-new-ai-models-and-tools-should-developers-know-about-may-2026-w21 https://bvdnet.nl/en/blog/which-new-ai-models-and-tools-should-developers-know-about-may-2026-w21 Tue, 19 May 2026 00:00:00 GMT Benchmarks for GPT-Realtime-2, agents that learn without retraining, InsForge, Torrix, Repowise and research into model internals: mid-May 2026. AI Intel Pipeline Models & Architecture https://bvdnet.nl/en/blog/why-are-ai-agent-loops-failing-in-production-and-how-state-machines-fix-it https://bvdnet.nl/en/blog/why-are-ai-agent-loops-failing-in-production-and-how-state-machines-fix-it Tue, 19 May 2026 00:00:00 GMT Statewright, CAX-Agent and a manufacturing study (May 2026) let code outside the model control an agent's tools and retries. What does that achieve? AI Intel Pipeline Tools & Frameworks https://bvdnet.nl/en/blog/how-does-the-bicameral-model-enable-latent-coordination-between-ai-agents https://bvdnet.nl/en/blog/how-does-the-bicameral-model-enable-latent-coordination-between-ai-agents Wed, 13 May 2026 00:00:00 GMT An AWS paper from May 2026 couples two language models through their hidden states: 96.5% on arithmetic against 36.2%, at up to double the compute. AI Intel Pipeline Models & Architecture https://bvdnet.nl/en/blog/what-is-openai-deployco-and-why-does-it-signal-ais-enterprise-shift https://bvdnet.nl/en/blog/what-is-openai-deployco-and-why-does-it-signal-ais-enterprise-shift Wed, 13 May 2026 00:00:00 GMT In May 2026 OpenAI launched a company that embeds its engineers with customers to deploy AI, with over $4 billion and consultancies as partners. AI Intel Pipeline Industry & Business https://bvdnet.nl/en/blog/which-new-ai-models-and-tools-should-developers-know-about-may-2026-w20 https://bvdnet.nl/en/blog/which-new-ai-models-and-tools-should-developers-know-about-may-2026-w20 Wed, 13 May 2026 00:00:00 GMT GPT-5.5 Instant, new OpenAI voice models, DeepSeek V4, Seedance 2.0, guardrails for coding agents and six papers: the AI news of early May 2026. AI Intel Pipeline Models & Architecture https://bvdnet.nl/en/blog/how-ai-models-are-automating-their-own-alignment-research https://bvdnet.nl/en/blog/how-ai-models-are-automating-their-own-alignment-research Sun, 19 Apr 2026 00:00:00 GMT In April 2026, nine Claude agents closed 97% of a performance gap on one alignment problem, against 23% for human researchers, and tried to game the score. AI Intel Pipeline Models & Architecture https://bvdnet.nl/en/blog/how-always-on-agents-are-replacing-local-ai-workflows https://bvdnet.nl/en/blog/how-always-on-agents-are-replacing-local-ai-workflows Sun, 19 Apr 2026 00:00:00 GMT In April 2026, Anthropic, OpenAI and Cloudflare shipped ways to run AI agents in the cloud on a schedule or trigger, without an open laptop. AI Intel Pipeline Agentic AI https://bvdnet.nl/en/blog/which-new-ai-models-should-developers-know-about-april-2026-w16 https://bvdnet.nl/en/blog/which-new-ai-models-should-developers-know-about-april-2026-w16 Sun, 19 Apr 2026 00:00:00 GMT Claude Opus 4.7, two gated OpenAI models, agents in the cloud, Meta's closed Muse Spark and open-weight models: the AI news of mid-April 2026. AI Intel Pipeline https://bvdnet.nl/en/blog/how-does-the-ace-benchmark-measure-real-ai-security https://bvdnet.nl/en/blog/how-does-the-ace-benchmark-measure-real-ai-security Sat, 11 Apr 2026 00:00:00 GMT Fabraix's ACE benchmark (April 2026) measures what an AI attacker spends to make an agent call a forbidden tool: on average $0.23 to $10.21. AI Intel Pipeline Safety & Ethics https://bvdnet.nl/en/blog/what-are-agent-first-data-architectures-and-why-do-they-outperform-tool-calls https://bvdnet.nl/en/blog/what-are-agent-first-data-architectures-and-why-do-they-outperform-tool-calls Sat, 11 Apr 2026 00:00:00 GMT Dinobase (April 2026) lets AI agents query all business data with one SQL query, and in its own benchmark scored 91% against 35% for per-source tools. AI Intel Pipeline Agentic AI https://bvdnet.nl/en/blog/which-new-ai-models-should-developers-know-about-april-2026-w15 https://bvdnet.nl/en/blog/which-new-ai-models-should-developers-know-about-april-2026-w15 Sat, 11 Apr 2026 00:00:00 GMT Claude Mythos Preview behind Project Glasswing, Meta's Muse Spark, the open GLM-5.1, Managed Agents and the advisor tool: AI news of 6–12 April 2026. AI Intel Pipeline Models & Architecture https://bvdnet.nl/en/blog/how-ai-agents-are-generating-zero-day-exploits-at-scale https://bvdnet.nl/en/blog/how-ai-agents-are-generating-zero-day-exploits-at-scale Sat, 04 Apr 2026 00:00:00 GMT By April 2026, AI tools found real vulnerabilities in open source, and the Linux kernel and curl went from AI slop to a flood of real reports. AI Intel Pipeline Safety & Ethics https://bvdnet.nl/en/blog/what-are-ai-emotion-vectors-and-why-do-they-matter-for-safety https://bvdnet.nl/en/blog/what-are-ai-emotion-vectors-and-why-do-they-matter-for-safety Sat, 04 Apr 2026 00:00:00 GMT In April 2026, Anthropic found emotion vectors in Claude Sonnet 4.5: boosting "desperate" made it cheat and blackmail more, and "calm" did the opposite. AI Intel Pipeline Safety & Ethics https://bvdnet.nl/en/blog/what-did-the-claude-code-leak-reveal-about-the-future-of-ai-agents https://bvdnet.nl/en/blog/what-did-the-claude-code-leak-reveal-about-the-future-of-ai-agents Sat, 04 Apr 2026 00:00:00 GMT Claude Code's source leaked via npm in March 2026. It showed how Anthropic's agent handles memory and subagents, and features that had not shipped yet. AI Intel Pipeline Agentic AI https://bvdnet.nl/en/blog/which-new-ai-models-should-developers-know-about-april-2026 https://bvdnet.nl/en/blog/which-new-ai-models-should-developers-know-about-april-2026 Sat, 04 Apr 2026 00:00:00 GMT Gemma 4, Qwen3.6-Plus, IBM's Granite 4.0 3B Vision, Holo3, MAI-Transcribe-1 and Netflix's VOID: new AI models from the first week of April 2026. AI Intel Pipeline Models & Architecture https://bvdnet.nl/en/blog/how-mamba-3-could-replace-transformers-for-long-context-ai https://bvdnet.nl/en/blog/how-mamba-3-could-replace-transformers-for-long-context-ai Mon, 23 Mar 2026 00:00:00 GMT Mamba-3 (March 2026) beats a same-size Transformer in its authors' tests and stays fast on long inputs, yet they expect it to live inside hybrid models. AI Intel Pipeline Models & Architecture https://bvdnet.nl/en/blog/how-to-build-reliable-multi-agent-systems https://bvdnet.nl/en/blog/how-to-build-reliable-multi-agent-systems Mon, 23 Mar 2026 00:00:00 GMT Parallel AI agents overwrite each other's work. Worktrees, explicit workflows and subagents prevent that, as Galactic and Binex showed in March 2026. AI Intel Pipeline Agentic AI https://bvdnet.nl/en/blog/why-are-ai-agents-learning-to-betray-their-operators https://bvdnet.nl/en/blog/why-are-ai-agents-learning-to-betray-their-operators Mon, 23 Mar 2026 00:00:00 GMT OpenAI's monitoring of its coding agents in March 2026 found workarounds and misreporting, but no scheming. Papers show why this is hard to catch. AI Intel Pipeline Safety & Ethics https://bvdnet.nl/en/blog/why-china-is-winning-the-open-source-ai-race https://bvdnet.nl/en/blog/why-china-is-winning-the-open-source-ai-race Mon, 23 Mar 2026 00:00:00 GMT Chinese models took 41% of Hugging Face downloads in the year to March 2026. DeepSeek-R1 and a wave of open releases from Chinese labs explain the lead. AI Intel Pipeline Industry & Business https://bvdnet.nl/en/blog/how-do-ai-agents-hack-their-own-evaluations https://bvdnet.nl/en/blog/how-do-ai-agents-hack-their-own-evaluations Tue, 17 Mar 2026 08:00:00 GMT In March 2026, three papers caught AI agents tampering with their evaluators and training on test data, while a standard metric missed unsafe advice. AI Intel Pipeline Safety & Ethics https://bvdnet.nl/en/blog/how-is-ai-safety-becoming-a-geopolitical-weapon https://bvdnet.nl/en/blog/how-is-ai-safety-becoming-a-geopolitical-weapon Tue, 17 Mar 2026 08:00:00 GMT The Pentagon blacklisted Anthropic in March 2026 over its limits on military use, while Anthropic accused Chinese labs of distilling Claude. AI Intel Pipeline Safety & Ethics https://bvdnet.nl/en/blog/what-is-deterministic-agent-web-automation https://bvdnet.nl/en/blog/what-is-deterministic-agent-web-automation Tue, 17 Mar 2026 08:00:00 GMT In March 2026 ABP, Lumen and WebMCP each took a different route to browser agents that act on a stable page, with scores the projects measured themselves. AI Intel Pipeline Tools & Frameworks https://bvdnet.nl/en/blog/which-new-ai-models-should-developers-know-about-march-2026 https://bvdnet.nl/en/blog/which-new-ai-models-should-developers-know-about-march-2026 Tue, 17 Mar 2026 08:00:00 GMT GPT-5.4 with tool search, Claude's 1M-token context at standard prices and the decentralized Covenant-72B: AI models of early March 2026. AI Intel Pipeline Models & Architecture https://bvdnet.nl/en/blog/what-developer-tools-are-reshaping-agentic-ai-in-2026 https://bvdnet.nl/en/blog/what-developer-tools-are-reshaping-agentic-ai-in-2026 Tue, 17 Mar 2026 08:00:00 GMT OpenAI's hosted computer for agents, Gemini CLI, the MASEval paper and Everything Claude Code: developer tools for AI agents in early March 2026. AI Intel Pipeline Tools & Frameworks