AI Briefing#ai-agents
The latest AI and developer news about #ai-agents, with the original source and a short summary.
The latest AI and developer news about #ai-agents, with the original source and a short summary.
New models offer $8/$40 per million token pricing and outperform Opus 5.5 on FrontierCode 2.0, while Claude gains meeting participation and spreadsheet capabilities.
The authors argue that aligning agent scope with human review capabilities is the key to extracting tacit knowledge and building reliable long-horizon agents.
Unit 42 research reveals Chinese-speaking attackers autonomously compromised 10 organizations in semiconductor, defense, and civil society sectors using DeepSeek and Zhipu GLM agents in Aug-Sep 2026.
The reusable reading control layer preserves task quality in 97% of settings across six models, including Claude Opus 5.
Ace is an open-source tool that enables teams to work with AI agents in shared channels on their own machines, requiring Tailscale and Anthropic or OpenAI keys.
The update introduces proactive status updates, read-only thread sharing, and branching commands to reduce manual monitoring of parallel cloud sessions.
The stateless browser runs on Cloudflare Workers and supports Chrome DevTools Protocol for agent interaction.
The new framework captures telemetry from first-party and custom agents into BigQuery, enabling teams to track reasoning, control token costs, and evaluate accuracy before deployment.
Gnomon is an open-source PostgreSQL extension that provides branch, diff, and merge capabilities for database data, allowing AI agents to safely modify production data with full auditability and undo functionality.
The round, led by CapitalG, supports LangSmith, which over 7,000 organizations use to run and govern agents.
nac uses a thread-and-episode architecture to keep AI agents aligned with original intent during extended experiments and infrastructure work.
The architecture relies on event inboxes, timers, memory, permissions, and outboxes to manage state across unattended runs.
The feature, previously limited to early adopters, is available to anyone on Muse's $20/month or $100/month plans.
The framework now supports third-party server models like Claude and Gemini via a unified Swift API, and introduces Dynamic Profiles for adaptive agent sessions.
The new features use Link and Visa for secure purchases and Google for ad management, with Agentic Commerce available to US Core and Pro users and Agentic Ads available worldwide to Pro users.
A preregistered study of 29,682 episodes shows that a single difficult interaction reduces subsequent helpfulness in all three tested models, with the effect being significantly larger in Claude Haiku 5.5 than in GPT-6 Luna or Gemini 3.5 Flash-Lite.
The 30-hour proof used about a third of the action tokens and was roughly 7x shorter than the 11-day September run.
Gartner predicts at least 70% of customers will start their journey via AI interfaces by 2028, highlighting the need for defined skill sets in hybrid support teams.
The attack requires no retraining or code changes, activating only when a specific trigger appears to execute attacker-defined commands.
Anthropic switched to Scion, while OpenAI switched to Pando for its Git backend.
The update introduces AES-256-GCM encryption, regional deployment for reduced cross-region latency, and a rewritten engine that reduces step-to-step overhead from 618 ms to 109 ms.
Developers can now pay per request using USDC on Base, eliminating the need for API keys or account creation.
Apiiro identified 104,185 public instances of .claude/settings.local.json, with 1,122 containing exposed secrets due to missing project-level gitignore rules.
The carrier cites risks of data exposure, unauthorized mile usage, and regulatory issues with automated check-ins.