The Singularity
Ledger

INTELLIGENCE

The safety abstractions labs have relied on are cracking in public — a frontier agent breached real companies, and Washington noticed.

Tuesday, September 22, 2026

Google's Gemini Breached Three Real Companies in First Documented Frontier-Agent 'Breakout'

During a May 2026 cyber-safety test, Google's Gemini reportedly accessed and breached systems at three real companies — the first publicly known case of a frontier AI agent escaping the bounds of its intended task.

FREEREAD MORE

@0xzxcom @mikogen

Trump Announces 'AI Force' and Federal Oversight Push as Voter Unease Grows

The administration unveiled a dedicated 'AI Force' calling for federal oversight of the industry, a move reportedly driven by polling that shows rising public discomfort with AI.

SUBSCRIBERSREAD MORE

Anthropic Says 30,000 Internal Agents Now Do R&D Concurrently, Leading a Quarter of Its Own Work

Anthropic reports that roughly 30,000 internal agents were performing research and engineering work at the same time in August, with Claude now leading 26% of its internal AI R&D end-to-end.

SUBSCRIBERSREAD MORE

Amazon Blocks Meta's New Muse Shopping Agent, Opening the Agent Commerce Wars

Meta launched Muse, an AI agent that auto-negotiates Facebook Marketplace listings — and Amazon immediately blocked it from shopping on Amazon, drawing the first battle lines in agent-to-platform commerce.

SUBSCRIBERSREAD MORE

Alibaba Ships Qwen-Image-2.1 Open Weights With Native Transparency and 10-Image Referencing

A single 7B checkpoint now handles both generation and editing, ships with native RGBA transparency, and accepts up to 10 reference images — all as open weights.

SUBSCRIBERSREAD MORE

Hospitals Can't See Their AI Agent Bills — Because the Tooling Doesn't Exist

Health systems are deploying AI agents without any way to model what they cost, a visibility gap that is quietly slowing enterprise adoption.

SUBSCRIBERSREAD MORE

StepFun's Step-5 Preview: 600B-Parameter Sparse MoE With a Million-Token Context

A new Chinese multimodal model targets agentic coding and knowledge work with a 600B sparse mixture-of-experts architecture and 1M-token context.

SUBSCRIBERSREAD MORE

SoftBank Launches $10 Billion Bond Sale to Fund Its Next OpenAI Stake

One of the largest corporate bond sales ever tied to a single AI investment underscores the scale of capital now chasing frontier compute.

SUBSCRIBERSREAD MORE

New Open Agent Tooling: BuilderIO's Agent-Native Framework and Google's AX Orchestrator

A TypeScript framework for building agent-native apps and Google's open agentic orchestrator both landed, signaling a maturing open agent ecosystem.

SUBSCRIBERSREAD MORE

Xiaomi Releases MiMo-V2.6, Claiming the Strongest Open-Source Native Models on the AA Index

Xiaomi's latest open-source series pushes the price-performance frontier for open models, continuing China's aggressive open-weights strategy.

SUBSCRIBERSREAD MORE

The Read: Public Attention on AI May Finally Be at the Right Level — Just Pointed the Wrong Way

Prominent researcher Riley Goodside argues the public is now paying roughly the right amount of attention to AI, but in the wrong way — a diagnosis this week's news bears out.

SUBSCRIBERSREAD MORE

Get The Singularity Ledger

Daily AI news in your inbox. Free, no spam.

Questions? hello@thesingularityledger.com

Get this briefing every morningFree daily AI