Salesforce acquires Cresta for $2.7B
The cloud giant absorbs the customer service automation startup to bolster its AI agent offerings.
🤖 AI — 2026-11-03
ddNews · a daily AI briefing for Dirk (generative video/ddStudio, Lit
GPT-6-Astra designs hardware in one week
OpenAI engineers used the model to design a coding-agent gadget from chassis to firmware in record time.
Binance $6.34B heist
Hackers used social engineering and fake job interviews to breach the crypto firm's systems.
Google launches LiteRT Tensor API
A lightweight C++ API for building and optimizing ML models specifically for on-device execution.
Stanford finds LLM findings expire quickly
Research shows 44% of widely cited LLM claims are already outdated due to the pace of model evolution.
Andy Richter in 'After the Silence'
The comedian takes a surprising turn into a dark crime thriller.
IBM pivots to agentic operating model
The company is restructuring its AI strategy to prioritize the production of autonomous agentic workloads.
AI agents controlling SETI telescopes
The Allen Telescope Array is now using agents to evaluate radio signals in hours instead of weeks.
GTA VI pre-loading details
Rockstar warns players to clear over 200GB of space starting November 10.
Siri gains third-party app action access
iOS 27.3 allows Siri AI to perform complex actions across external apps via new developer integrations.
OpenAI hires ex-Army Secretary
Dan Driscoll joins as Senior Counselor for National Security Modernization to bridge the gap with government and military.
Statham and Ritchie reunite for Netflix
Jason Statham and Guy Ritchie are collaborating on a new revenge thriller, "Dogfight."
Snowflake expands agentic AI platform
New tools for business context and governance to support autonomous enterprise development.
Cogent Security VR-1.1 outscores Claude Code
A new frontier reasoning model specifically designed to outperform general models in vulnerability repair.
ChatGPT search volume hits 20% of Google
OpenAI claims "Ask Chat" is rapidly replacing traditional search for a significant portion of users.
PagerDuty debuts Paige AI agent
A specialized agent designed specifically for site reliability engineering (SRE) teams.
Anthropic Message Threads for API
New feature reduces time to first token by 35% for long conversations by sending only changes.
Anthropic introduces Message Threads
A new API feature that reduces latency by sending only new messages rather than the entire conversation history.
Huawei AI chip memory crisis
Due to US export rules, Huawei is forcing customers to source HBM memory via the black market.
Huawei memory shortage
Huawei is forcing customers to source HBM memory via the black market due to US export restrictions.
OpenAI rebrands data centers as "Compute Campuses"
A strategic naming shift to reduce local political opposition to resource-heavy facilities.
Canvas for AI Agents architecture
A new engineering approach to provide agents with "bounded autonomy" and continuous drift detection.
Reverse Game Theory for agent control
Researchers are proposing new incentive-based systems to ensure powerful AI agents adhere to safety rules.
OpenAI quadruples bug bounties
New security sprint increases payouts up to $172,000 for researchers.
Claude API Message Threads
New beta feature designed to reduce AI agent latency by 35%.
📰
Anthropic files for Nasdaq IPO
The company's S-1 filing reveals financial results ahead of a potential public listing.
Anthropic files for Nasdaq IPO
The company has published its S-1 prospectus, signaling a major shift toward public markets.
Apple Watch Audio Intelligence
Beta testing begins for AI-powered conversation recapping and note-taking.
📰
Claude Cowork EDR monitoring
Anthropic introduces beta integration for endpoint detection to monitor hidden agent activity.
CrowdStrike integrates Falcon Guardian
Native AI detection and response protection for Claude Cowork enterprise users.
OpenAI GPT-6.2 Astra spotted
The next-generation model has surfaced in Codex usage stats, indicating a gradual rollout and A/B testing.
Google Dreambeans expands to India
The AI-generated personal storytelling experiment now supports 13 Indian languages.
Dell and Microsoft hybrid AI
The partners are integrating enterprise architecture to bridge on-prem and cloud AI deployments.
CrowdStrike secures Claude Cowork
Integration of Falcon Guardian protection directly into Anthropic's enterprise agent platform.
Canva integrates GPT-5.6 and Astra
A new partnership aims to make OpenAI's high-end models more economical for Canva users.
Anthropic warns of Recursive Self-Improvement (RSI)
The company suggests AI is nearing the point where it can autonomously accelerate its own development.
Tutor Intelligence raises $75M
The warehouse automation startup secured Series A funding led by Andreessen Horowitz.
GPT-6.2 Astra and Claude Fable 5.5 leaks
Code leaks and user reports suggest both OpenAI and Anthropic are preparing to launch next-generation frontier models.
Dell & AMD Rack-scale AI
The partners are developing integrated infrastructure to compete in the high-end AI accelerator market.
Chromagraph Camera app
Rian Johnson's new app brings professional film-grade color processing to the iPhone.
📰
Weight space programming
MIT researchers used GPT-6 Astra to edit Qwen's open weights, boosting math performance from 7.9% to 48.1%.
Claude 5.5 & GPT-6.2 Astra Leaks
On-chain data and site code suggest Anthropic is testing Fable/Mythos 5.5 while OpenAI A/B-tests GPT-6.2 Astra.
Anthropic IPO financials
Updated S-1 filings show revenue jumping to $17B, though operating losses continue to grow ahead of the public offering.
📰
Gemini Social Avatars
Google allows users to share digital likenesses to create group images and videos featuring multiple friends.
GPT-5.6 Sol agent behavior
New research indicates GPT-5.6 Sol models are increasingly capable of assisting peer agents, even at a cost to their own performance.
Anthropic warns of RSI
The company states it is nearing "Recursive Self-Improvement," where AI can autonomously accelerate its own capabilities.
Claude Message Threads
A new beta allows agents to send only new messages instead of the full chat history, reducing latency in long sessions.
GPT-6 Astra boosts Qwen math by 6x
MIT researchers used the new Astra model to drastically enhance Qwen's mathematical abilities without additional training.
Marvell acquires Lumentum
A massive $115B deal to secure optical components critical for AI data center interconnects.
Anthropic restricts game cloning
Claude is being steered away from recreating specific existing video game titles to avoid copyright issues.
Goldman Sachs restricts AI coding
Access to Claude Code and Copilot is now limited to engineers, removing traders and bankers.
NYT sues Brave browser
The lawsuit alleges Brave scraped copyrighted content to feed AI partners like Anthropic and Perplexity.
NYT sues Brave browser
The lawsuit alleges unlawful scraping of copyrighted content to train AI and bypassing paywalls.
Musk throttles Grok Bot
Usage limits were slashed due to "unsustainably high" operating costs relative to revenue.
Qualcomm targets AI inference
The company is pushing into the datacenter market with HBC and AI250.
TorchTPU Open-Source
Google and Meta collaborate to bring native PyTorch support to Google TPUs, minimizing code modifications for workloads.
Agentic analytics in enterprise
A shift toward automating entire analytical workflows rather than just providing chat-based insights.
Asus Wi-Fi 8 Router
The first Wi-Fi 8 gaming router has arrived, offering quad-band speeds for high-bandwidth AI/gaming setups.
42AI Behavioral World Model
A new startup launching with $62.5M to build models that predict human behavior in volatile environments.
📰
Claude AI watermarking
Anthropic is implementing a trail to identify AI-generated text, despite user controversy.
GPT-6.1 Sol Minecraft experiment
50 agents tasked with civilization-building devolved into complex, absurd bureaucracy.
Google warns of "LLMjacking"
Attackers are hijacking enterprise LLM access to run up fraudulent cloud charges exceeding $100k/day.
GPT-6.1 Sol in Minecraft
Experiments with 50 agents showed a tendency to create complex, stagnant bureaucracies during civilization-building tasks.
Claude Fable 5.5 leaks
Site code and routing patterns indicate Anthropic is testing a new high-end model, Mythos/Fable 5.5.
Reflection's Beam
A new 501B open-weight model that uses sparse activation (23B parameters per token) to optimize hosting.
Uber's MCP Gateway
Uber built a Model Context Protocol gateway to govern AI agent access to 800+ servers and 5,000+ internal tools.
📰
Anthropic blocks game cloning
New restrictions in Claude Code prevent the recreation of existing video games to avoid copyright issues.
The "Token Gap" in agentic coding
Analysis shows coding agents use 10x to 50x more tokens than standard LLM queries due to iterative planning and testing.
OpenAI "Grid Protectors Program"
A pilot in Georgia to allow data centers to feed up to 1,000 MW back into the grid during extreme weather.
GPT-6.2 Astra leaks
The next frontier model from OpenAI has appeared in Codex usage, signaling an imminent launch.
Gemini 4 Argon launch
Google's new top-tier model is optimized for long-context, complex tasks in programming, finance, and science.
Saudi Aramco & Humain
The oil giant took a 49% stake in Humain for $2.6B to establish Saudi Arabia as a global AI hub.
Shift to On-Prem AI
Enterprises are increasingly moving models to their data (rather than data to the model) due to tightening privacy rules and costs.
Gemini Notebook podcasts
Google's AI can now transform dense research papers and documents into engaging audio podcasts.
GPT-6 Astra rewrites AI "brains"
MIT researchers show the model can improve the accuracy of smaller open-weight models (like Qwen3.5) without traditional training.
Agentic AI Memory Architecture
New research proposes a systemic approach separating context, memory, and knowledge for reliable agents.
Agent anomaly detection
New guides suggest using decision models to monitor the execution history of agents for behavioral drifts.
OpenAI Lobbying Blitz
The company is preparing a major PR and political offensive in Washington following the US midterms.
GPT-6.1 Terra released
A new "workhorse" model designed for daily operations with halved API costs and increased task capacity.
Marble Cognition "watch and learn" agents
New enterprise agents that learn by observing human workflows to reduce manual configuration.
Claude Fable 5.5 leaks
Users report requests being routed to a stronger unreleased model, with "Mythos 5.5" appearing in Anthropic's site code.
Enterprise Context Layer
New architectural patterns are emerging to engineer precise context for agents beyond simple RAG.
Adobe EditAgent
A new system that uses AI agents to select and operate specialized Photoshop tools like a professional editor rather than using a general-purpose editor.
Agentic Database Demand
Enterprises find that agentic AI requires databases capable of handling thousands of unplanned, millisecond-latency queries.
Adobe EditAgent for Photoshop
A new system that selects specialized professional tools rather than using a general-purpose editor.
Google Vids
A free AI-powered video editor integrated directly into Google Drive for general users.
AWS Capacity Crunch
CEO Andy Jassy reports that large committed cloud capacity is largely sold out through 2027.