Presented by
Good morning, AI enthusiasts. An engineer at Meta posted a question on an internal forum. An AI agent responded with what looked like a good solution. The engineer followed it. Two hours later, sensitive user and company data had been exposed to Meta engineers across the company.
In today's recap:
Meta's AI agent exposes sensitive internal data
OpenAI acquires Python toolmaker Astral for Codex
Jeff Bezos targets $100B for AI manufacturing fund
4 new AI tools, prompts, and more
META
Meta's AI agent helped, then leaked data
Recaply: Meta just confirmed that one of its internal AI agents caused a large-scale data leak, exposing sensitive user and company data to engineers for two hours after responding to a routine internal forum query.
Key details:
An employee asked the AI agent for guidance on an engineering problem via an internal forum; the agent's suggested solution, when implemented, inadvertently exposed the data to Meta engineers across the company.
The exposed data covered both sensitive user and company information and remained visible to engineers for two hours, with Meta describing it as "a major internal security alert."
Consulting firm co-founder Tarek Nseir told the Guardian: "If you put a junior intern on this stuff, you would never give that junior intern access to all of your critical severity one HR data," adding the vulnerability should have been obvious in retrospect.
Meta confirmed the incident on March 20, saying "no user data was mishandled" and adding that a human engineer could also give erroneous advice; the incident was first reported by The Information.
Why it matters: AI agents don't carry the context a veteran engineer picks up over years. An engineer knows which systems touch customers and what breaks at 2am. AI agents don't have that memory. Meta's incident is part of a growing pattern. Amazon had two AI outages in February after similar deployments went wrong. More will happen. The answer isn't to slow down on agents. It's to build proper access controls before deploying them near sensitive data.
PRESENTED BY ATTIO
Here’s how I use Attio to run my day.
Attio is the AI CRM with conversational AI built directly into your workspace. Every morning, Ask Attio handles my prep:
Surfaces insights from calls and conversations across my entire CRM
Update records and create tasks without manual entry
Answers questions about deals, accounts, and customer signals that used to take hours to find
All in seconds. No searching, no switching tabs, no manual updates.
Ready to scale faster?
OPENAI
OpenAI buys python toolmaker Astral for Codex
Recaply: OpenAI just agreed to acquire Astral, the maker of uv, Ruff, and ty, bringing the most widely used Python developer tools into its Codex ecosystem as it expands beyond code generation into the full development workflow.
Key details:
Astral's tools sit at the core of modern Python development: uv manages dependencies and environments, Ruff handles linting and formatting at high speed, and ty enforces type safety across codebases, with all three remaining open source after the acquisition.
Codex now has 2 million weekly active users and has seen 3x user growth and 5x usage increase since the start of 2026, making it one of the fastest-growing developer platforms in AI right now.
Charlie Marsh, Astral's founder and CEO, said the team will "continue evolving our open source tools to push the frontier of software development" after joining Codex; OpenAI also acquired Promptfoo earlier in March 2026.
The acquisition is subject to regulatory approval; until closing, OpenAI and Astral remain separate and independent companies.
Why it matters: OpenAI doesn't just want to write your code anymore. The company wants to own the entire workflow that developers use to build software from start to finish. Anthropic built Claude Code and Cursor built a frontier model of its own. Now OpenAI owns the Python tools that millions of developers rely on every single day. The race for developer tooling is speeding up fast, and every big AI lab is moving to sit deeper in your stack. For developers, picking tools is now a strategic bet.
ENTERPRISE AI
Bezos yes $100B to remake manufacturing with AI
Recaply: Amazon founder Jeff Bezos is reportedly in talks to raise $100B for an AI manufacturing fund linked to Project Prometheus, targeting acquisitions of legacy manufacturers in sectors like chips and defense for AI-driven transformation.
Key details:
Project Prometheus, an AI startup where Bezos serves as co-CEO, aims to apply AI to engineering and manufacturing operations, with the fund designed to buy companies outright and rebuild their workflows using AI technology.
Project Prometheus previously raised $6.2B in late 2025, according to a Financial Times report from February; the new $100B target would be more than 16x that initial raise and rival the scale of the Stargate initiative.
Bezos has already traveled to the Middle East and Singapore in active fundraising efforts for the initiative, according to the Wall Street Journal, signaling serious international capital interest in the fund.
Fundraising conversations are described as early stage; Project Prometheus was first reported publicly in late 2025 when the company confirmed it had $6.2B in committed funding.
Why it matters: The biggest AI bets so far have gone into data centers and compute. Bezos is targeting something different: the physical economy. Project Prometheus wants to buy manufacturing companies and remake them with AI. If the $100B raise closes anywhere near its target, it would be the biggest single AI bet aimed at the real economy rather than software. The arms race has moved beyond chips. It's now about who owns the factories.
TOOLS
Trending AI Tools
🎨 MAI-Image-2 - Microsoft's new text-to-image model
🎬 Original Series - Higgsfield's AI-native streaming platform
📱 DoorDash Tasks - DoorDash's new app that pays couriers to submit videos to train AI
🚀 AI Studio - Google’s upgraded full-stack vibe coding agent
NEWS
What Matters in AI Right Now?
Cursor launched Composer 2, a frontier-level coding model priced at $0.50/M input and $2.50/M output tokens, scoring 73.7 on SWE-bench Multilingual and 61.7 on Terminal-Bench 2.0 through its first continued pretraining run.
Google introduced a full-stack vibe coding experience in AI Studio, powered by the new Antigravity coding agent with Firebase integration, supporting multiplayer experiences, external libraries, and secure login without leaving the platform.
Anthropic released Claude Code Channels in research preview, letting users control Claude Code sessions via Telegram or Discord MCPs and send coding instructions directly from their phone.
The co-founder of Supermicro and two associates were charged with diverting $2.5B in Nvidia-equipped servers to China through ASEAN shell companies using dummy servers and fake labels, with Super Micro shares falling 14% following the arrests.
Lovable announced an expansion beyond app building, with its platform now positioning as a general-purpose AI co-founder covering data analysis, marketing assets, pitch decks, and business analytics alongside full-stack app development.
Alibaba's workforce shrank 34% in 2025 as the Chinese tech giant doubled down on AI, with net profit falling 66% year-over-year in Q4 2025 due to heavy AI investment spending.
OpenAI confirmed it will merge its ChatGPT app, Codex coding platform, and Atlas browser into a single desktop superapp, with head of applications Fidji Simo leading the effort to consolidate and simplify its product lineup.
Nvidia secured a deal to supply 1 million AI chips, networking equipment, and inference hardware to Amazon Web Services by end of 2027, with both NVDA and AMZN stocks rising following the announcement.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
