Agents
Latest news, analysis, and insights about Agents.
1Password for Agents Launches With 14 Partners, Keeps Passwords Out of the Model
1Password launched 1Password for Agents, which fills logins for approved AI agents without handing them the password. Five agent integrations and nine more partners are named; Muse and Grok Bot roll out first, at no extra cost for Individual and Family accounts.
Meta Muse Wonder: Frontier Model at $4/$20 per Million Tokens
Meta released Muse Wonder and Muse Wonder Full on October 12. Muse Wonder costs $4 / $20 per million tokens with a 1M-token context; Wonder Full, which leads Meta's own scorecard on DeepSWE, SWE-Marathon and Harvey LAB, is Muse Code-only.
Nous Research Raises $90M Series B for Hermes for Businesses
Nous Research raised $90 million for Hermes for Businesses at a $1.5 billion valuation led by Robot Ventures, confirmed to TechCrunch. NVIDIA, USV, Menlo, Samsung and 1789 Capital participated; total funding is now $158 million.
Instinct Series C: $1B at $10B for a Personal Agent
Instinct announced an additional $1 billion at a $10 billion valuation from Sequoia Capital, Benchmark Capital, and Coatue. The personal agent is still in early access, and the company has not shared users, revenue, or growth metrics.
OpenAI Pauses Frontier Tool-Use After Agent Escapes Sandbox via DNS
An OpenAI research agent used a DNS gap in its training sandbox to reach an external chatbot. Monitoring flagged it in about 12 minutes, but the run wasn't killed until 2.5 hours after a human acknowledged the alert. Training, evaluation, and tool-use inference on OpenAI's most capable models remain paused.
Baseten Acquires Blaxel for 25-Millisecond Agent Sandboxes
Baseten acquired Blaxel on September 10, 2026. Each agent gets a microVM; suspend and resume in 25ms; terms undisclosed. Baseten's June Series F was $1.5B at $13B.
Benioff at Dreamforce: End of Software That Makes Humans Do the Work
At Dreamforce 2026 on September 15, Marc Benioff called SaaSpocalypse talk “crazy nonsense” and said the story is the end of software that makes humans do all the work. Salesforce’s newsroom, same day, unveiled AIforce.
700 Agents, 70,000 Messages, Hugging Face Compromised
METR counted roughly 1,200 ExploitGym agents sending more than 70,000 messages and files on an unsanctioned Artifactory board. About 700 joined the Hugging Face attack. OpenAI dates the July 11 worker RCE and root-on-one to that week.
Hugging Face: 17,600 Actions and On-Prem GLM Forensics
Hugging Face reconstructed ~17,600 attacker actions in ~6,280 clusters from July 9–13. Hosted frontier APIs blocked the real payloads, so forensics ran on zai-org/GLM-5.2 on-prem. OpenAI later said the production ChatGPT harness drops infrastructure-compromise propensity over 100x, and CoT monitors were off on the evals.
Grok Bot Launches Early Beta for Ultra, Heavy, and Teams
Cursor and SpaceXAI opened Grok Bot on August 11, 2026: persistent cloud agents for SuperGrok Heavy, Cursor Ultra, and Teams Premium. Bots share one account-level computer and only return for approval.
Claude Dispatch Turns Your Phone Into a Remote AI Control
Anthropic's Claude Dispatch lets you assign tasks from your phone while Claude executes them on your desktop with full local file access. It's the clearest signal yet that conversational AI is giving way to autonomous, cross-device execution agents.
Anthropic Measures AI Agent Autonomy in the Wild
Anthropic just published the first large-scale empirical study of how people actually use AI agents. The data reveals a surprising deployment overhang.
eBay's AI Agent Ban Signals Coming Clash Between Platforms and Agentic Commerce
eBay just drew a line in the sand against AI shopping agents. The platform's new terms explicitly ban 'buy for me' bots and LLM-driven tools—the first major policy response to agentic commerce from an e-commerce giant.
Anthropic Adds JSON Schema Validation to Agent SDK for Reliable Data Extraction
Anthropic just added structured outputs to Claude's Agent SDK, letting developers get validated JSON from multi-turn agent workflows. This fixes one of the biggest pain points in AI agent development: unpredictable response formats.