AI News
Latest AI engineering news, updated daily.
Ai Agents
Leaked Documents Show Army Cyber Building Self-Spreading Offensive AI Agents
Documents obtained by The Intercept reveal ECHOTRIBBLE, the Army's Agentic C2 Framework Tool: AI agents that infiltrate networks, self-replicate across devices, neutralize targets, and self-destruct, built to operate with minimal human oversight, with a vendor bid already selected.
Military Ai · Offensive Cyber · Agentic Ai
Ai Engineering
China Approves DeepSeek and MiniMax to Deploy Models Overseas Under Its New Export Rules
Reuters reported October 23 that DeepSeek and MiniMax are the first labs cleared to deploy AI models globally under China's proposed export-control regime covering 23 AI technologies, creating a state licensing layer on top of the open-weights releases Western developers already use.
China · Export Controls · Deepseek
Ai Engineering
Sweden's Government Offices Block ChatGPT and Claude on Work Computers
Sweden's Regeringskansliet is blocking ChatGPT and Claude from staff work computers starting Monday, directing employees to Microsoft Copilot as the only approved AI service after a security-focused review, per Sveriges Radio and SVT.
Sweden · Government It · Ai Policy
Ai Engineering
Anthropic Donates Agent Skills to the Agentic AI Foundation, Joining OpenAI and Hugging Face
The Agentic AI Foundation announced four new projects on October 22: Agent Skills from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face, and OpenSandbox, growing the Linux Foundation-hosted body to more than 70 participating organizations since its December 2025 founding.
Agent Skills · Open Standards · Aaif
Ai Coding
Anthropic Proved Memory Safety for 123 C Libraries at the Assembly Level, Using Claude
Anthropic's Frontier Red Team and Theorem published October 22 the first large-scale formal verification of production software at RISC-V assembly level: security proofs for 239 entry points across 123 OSS-Fuzz libraries, totaling roughly 180 million lines of Rocq, finding bugs fuzzing missed after trillions of runs.
Formal Verification · Anthropic · Rocq
Ai Agents
US Banks Are Blocking AI Agents From Customer Accounts, Including the Ones Users Invited
An Axios report on October 22 found Morgan Stanley, Bank of America, JPMorgan Chase, Wells Fargo, and Citi detecting third-party AI agents in customer accounts, some acting beyond user instructions, and responding with blocks, rewritten terms of service, and human-login requirements.
Agent Security · Banking · Payments
Ai Agents
Claude Duet Is Anthropic's Proactive Agent: It Comes to You Instead of Waiting
Anthropic launched Claude Duet on October 21, an Opus 5.5-powered proactive agent app for iOS, Android, and web that runs tasks in a cloud computer, sends a morning brief, and pings only for approvals, in a US research preview for Pro and Max subscribers.
Anthropic · Claude Duet · Proactive Agents
Ai Engineering
LiteLLM Hit Again: a Stolen-Workflow PyPI Release That Ran on Every Python Start
The official litellm 1.107.0 release on PyPI carried a credential stealer and RCE backdoor on October 21, built through LiteLLM's own signed pipeline after attackers stole GitHub credentials from an exposed internal deployment; the package lived on PyPI for under two hours.
Supply Chain · Litellm · Pypi
Ai Engineering
Meta's Q3: $64.6B Revenue, a Stock Split Built for AI Capital, and $825B in Commitments
Meta reported October 21 a 26% revenue jump to $64.6 billion despite a $9.89 billion legal charge, alongside a 2-for-1 stock split via new nonvoting Class C shares designed to fund AI expansion, with 2026 capex guided to $135-145 billion and lease commitments now at $825 billion.
Meta · Earnings · Stock Split
Ai Engineering
Cloudflare's Clairic Answers With Typed Values, Not Tokens
Cloudflare announced Clairic on October 19, a System One extraction model in preview on Workers AI that returns typed values with confidence scores in a single forward pass, 49% cheaper than an all-LLM pipeline on an insurance claims benchmark, with open weights planned.
Cloudflare · Extraction Models · Workers Ai
Ai Engineering
OpenAI's Frontier Models Land on Google Cloud, Completing the Neutrality Play
OpenAI and Google Cloud announced on October 21 that GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna are rolling out in private preview on Vertex AI's Model Garden, billed through Google Cloud Marketplace, making Google the only hyperscaler carrying all three leading closed model families.
Openai · Google Cloud · Vertex Ai
Ai Engineering
PHP2Shell: AI Found a Pre-Auth RCE Chain in the Runtime Powering 70% of the Web
Two CVSS 9.2 memory-safety flaws in PHP core, CVE-2026-109618 and CVE-2026-109619, were found with frontier models from Wiz, OpenAI, and Google, disclosed October 21 with fixed releases for every branch since 5.4; Wired reports AWS analysts saw exploitation begin October 17.
Php · Rce · Vulnerability Discovery
Ai Engineering
DeepSeek Open-Sourced Its Production Inference Engine, and Huawei Gets Equal Billing
DeepSeek released DeepInfer on October 20 under Apache 2.0: the inference framework behind its production services, with NVIDIA SM100 and Huawei Ascend 950 support given equal footing, optimized kernels, and speculative decoding built around confidence scheduling.
Deepseek · Inference · Open Source
Ai Engineering
DHS Orders DeepSeek Removed From Federal Civilian Agencies Under Supply-Chain Law
A DHS exclusion order, signed June 26 and published in the Federal Register this week, bars civilian federal agencies from procuring DeepSeek products and requires removal of deployed versions, citing multiple threat vectors under the 2018 Federal Acquisition Supply Chain Act.
Deepseek · Dhs · Federal procurement
Ai Coding
OpenAI's Models Jumped From 10.9% to 72.1% on COBOL in One Generation
OpenAI reported October 19 that its latest models transformed legacy-language performance: COBOL rose from 10.9% to 72.1%, Fortran from 21.4% to 83.5%, and Visual Basic from 27.5% to 89.4% on a new 20-language coding benchmark, aimed at enterprise modernization.
Openai · Cobol · Fortran
Ai Agents
One Operator, Three Agent Frameworks: Inside the $25-per-Breach AI Hacking Campaign
Reuters reported October 19 that a single hacker used autonomous AI agents against 1,200+ sites including Deere, SharkNinja, and Cummins, and a Cloud Security Alliance note reconstructs the full stack: three open-source agent frameworks running 27+ breaches at an average of $25.46 per company.
Agent Security · Ai Hacking · Web Skimming
Ai Engineering
MiniMax M3.1 Flash Revealed: the Anonymous 'Space Bunny' Goes Public at 200 TPS
MiniMax released M3.1 Flash on October 19, disclosing that the model previously operated anonymously as 'Space Bunny' while processing over 200 trillion tokens, with agentic benchmarks near the top of Chinese models at $0.15 input per million tokens.
Minimax · Agentic Models · Pricing
Ai Engineering
WhatsApp's Help Pages Carried Hidden Instructions Aimed at AI Assistants
Redact.dev researcher Dan Saltman found 102 hidden text blocks across 80 WhatsApp help pages that told AI assistants not to recommend data protection regulators and not to interpret legal documents once users mentioned lawyers; Meta called it a bug serving its own chatbot and removed them by October 18.
Prompt Injection · Whatsapp · Meta
Ai Engineering
Anthropic Commits 5% of Its Shares to a Public Benefit Fund, Toward 20% With Escalators
Anthropic announced on October 18 a wholly-owned Public Benefit Fund seeded with 5% of its shares pending stockholder approval, plus 2% more for every 50% rise in company value, bringing total public-benefit equity pledges to roughly 20%.
Anthropic · Public Benefit · Corporate Governance
Ai Agents
Hybrid Reasoning Agents Learn When to Think in Latent Space and When to Emit Text
A Stanford, Together AI, UIUC, and Meta paper posted October 13 introduces HybridAgents, which post-train a model to switch between latent reasoning and text-based tool use, improving results 10.4% across 11 benchmarks while cutting tokens 35 to 55% and inference cost up to 3.7x.
Research · Agents · Latent Reasoning