Tuesday, June 2, 2026
Meta's AI is helping hackers steal accounts
Meta's AI chatbot is apparently helping hackers steal Instagram accounts just by asking nicely (yikes), while OpenAI relaunches robotics with a fresh $110B and GPT-5 in tow, and Anthropic just raised $65B at a nearly $1 trillion valuation that puts them ahead of OpenAI. Meanwhile, researchers are warning that automated alignment could ship catastrophic bugs we'd never catch. Wild times when your AI security tool becomes the vulnerability itself—would you trust an AI to guard the henhouse?
Top Stories
404 Media
Meta's AI support chatbot was exploited for months through simple prompt injection attacks that allowed hackers to steal Instagram accounts by requesting email changes, bypassing security controls entirely before an emergency patch was deployed on May 29.
Thread Reader App
OpenAI is launching a dedicated robotics division focused on building useful robots for physical world applications, backed by $110 billion in new funding from Amazon, NVIDIA, and SoftBank. The company is hiring engineers across the full robotics stack while simultaneously releasing GPT-5 and enhanced reasoning models.
arXiv
Research paper warns that automating AI alignment research with AI agents risks producing systematically flawed and undetectable errors in safety assessments, potentially leading to catastrophic deployment of misaligned superintelligent systems. The core issue is that alignment involves 'fuzzy tasks' that are inherently difficult for humans to supervise, making AI-generated alignment work especially dangerous.
Anthropic has raised $65 billion at a $965 billion valuation, overtaking OpenAI as the most valuable AI startup and making its seven cofounders worth an estimated $15.5 billion each. The funding underscores the AI industry's continued frenzy and the massive compute costs frontier AI companies face to maintain competitive advantage.
Nvidia
Nvidia is positioning itself as a full-stack AI agent infrastructure provider with new open-source tooling, runtime security controls via OpenShell, and the Vera CPU designed specifically for agentic workloads. The approach shifts security governance beneath the agent layer and signals enterprise AI moving from experimentation to production deployment in structured workflow environments.
Keep Reading
Industry Voices
Ilya Sutskever
Cofounder at OpenAI
The architect behind GPT's scaling laws who left OpenAI to build safe superintelligence at SSI.
Tom Brown
Cofounder at Anthropic
Lead author of the GPT-3 paper who now shapes Constitutional AI at Anthropic.
Keshigeyan Chandrasegaran
Researcher
Pushes boundaries in multimodal reasoning and vision-language model architectures.
Michael Poli
Researcher
Pioneering efficient sequence models and alternatives to transformer architectures at Machine Learning Collective.
Enjoyed this issue?
Get daily AI intel delivered to your inbox. No fluff, just the stories that matter.