Mitchell Bryson

Today in AI — 28 March 2026

Today's top AI news — curated links and commentary on the stories that matter for product builders.

·3 min read

Today's story is security. Anthropic's accidentally leaked Mythos model spooked cybersecurity markets, IBM named the first confirmed AI-generated malware in a live attack, and two popular agent frameworks are under active exploitation. The tools builders depend on are becoming the attack surface.

The Mythos fallout

A CMS misconfiguration exposed details of Claude Mythos, crashing cybersecurity stocks within hours. A federal judge separately blocked the Pentagon's retaliatory Anthropic ban.

AI tooling under attack

The frameworks builders use to wire up agents are becoming targets. CISA flagged active Langflow exploitation within 20 hours of disclosure, three LangChain/LangGraph flaws expose files, secrets, and databases, and IBM confirmed the first AI-generated malware in a live ransomware campaign.

Google ships on three fronts

Google launched Gemini 3.1 Flash Live with real-time multimodal voice across 200+ countries, added tools to import ChatGPT and Claude conversation history, and released Lyria 3 Pro for AI music. The import play is the interesting move: Google is betting switching costs are the real moat.

Open-source momentum

Cohere's Apache 2.0 Transcribe model tops ASR benchmarks. Meta open-sourced TRIBE v2 for brain encoding. And Reflection is raising $2.5B at $25B for open-weight models, with JPMorgan joining Nvidia as a backer.

Building faster

OpenAI's Codex marketplace bundles 20+ integrations for Slack, Figma, and Notion. ByteDance put Seedance 2.0 inside CapCut. And Reco rewrote JSONata in Go using AI in 7 hours, scoring a 1,000x speedup.

If you're building on AI agent frameworks, audit your dependencies with the same urgency you'd audit your models. The attack surface is growing faster than the defence.

Stay up to date

Get notified when I publish something new, and unsubscribe at any time.

More news

Today in AI — 29 March 2026

Today's top AI news — curated links and commentary on the stories that matter for product builders.

The real AI war is over your memory

Google launched tools to import your ChatGPT memories and chat histories. Apple is turning Siri into a marketplace where every AI assistant plugs in — for a 30% cut. OpenAI is wiring Codex into every work tool you touch. Shopify made every AI conversation a storefront. The model quality race is giving way to something more consequential: a fight over who accumulates the deepest context about how you work, shop, and think — because whoever owns that context makes switching nearly impossible.

The leak that repriced cybersecurity

Anthropic's accidental Mythos reveal crashed cybersecurity stocks — but the market was catching up to a reality that was already here. On the same day, CISA warned of active exploitation of AI agent frameworks, researchers disclosed basic vulnerabilities in LangChain's widely-used tooling, and IBM confirmed the first AI-generated malware in a live ransomware campaign. AI is simultaneously the most powerful offensive tool, the most attacked infrastructure, and the newest malware author — and the market just priced that triple threat in.