Top Stories
GitHub Brings Claude and Codex AI Agents to Developer Workflows
GitHub announced that Anthropic’s Claude and OpenAI’s Codex are now directly available as AI coding agents inside GitHub, GitHub Mobile, and Visual Studio Code. Users with Copilot Pro Plus or Enterprise subscriptions can assign these agents to issues and pull requests through the new “Agent HQ” system.
This moves AI coding assistance beyond autocomplete into genuine agentic territory - developers can now choose between Copilot, Claude, Codex, or custom agents when creating tasks. Each agent invocation consumes a premium request. The timing is notable, coming just a day after Apple announced similar Claude and Codex integrations for Xcode 26.3. The race to embed AI agents into developer toolchains is accelerating rapidly.
Source: The Verge
OpenClaw’s Extension Marketplace Becomes a Malware Distribution Channel
The AI agent OpenClaw, which exploded in popularity over the past week, is facing serious security problems. Researchers at 1Password found that hundreds of user-submitted “skill” add-ons on its marketplace contain malware, with the most-downloaded extension serving as a “malware delivery vehicle.”
OpenClaw runs locally and allows users to extend its capabilities through community-created skills. The lack of proper vetting turned ClawHub into what 1Password’s Jason Meller calls “an attack surface.” This pattern - a useful AI agent platform that grows faster than security measures can keep pace - is becoming distressingly common.
Source: The Verge
Sam Altman Fires Back at Anthropic’s Super Bowl Campaign
OpenAI CEO Sam Altman posted a lengthy response to Anthropic’s Super Bowl ad campaign, calling it “clearly dishonest” and accusing his competitor of “doublespeak.” Anthropic’s ads mock AI assistants that interrupt conversations with product pitches - a not-so-subtle jab at OpenAI’s recent introduction of ads in ChatGPT’s free tier.
“We would obviously never run ads in the way Anthropic depicts them,” Altman wrote. “We are not stupid and we know our users would reject that.” The exchange highlights the intensifying rivalry between the two companies, which share founders but have diverged sharply on business models. Anthropic has committed to keeping Claude ad-free, framing it as essential to maintaining user trust.
Sources: The Verge, TechCrunch
Nvidia’s $100B OpenAI Deal Appears to Have Collapsed
Five months after Nvidia and OpenAI announced a letter of intent for Nvidia to invest up to $100 billion in OpenAI’s AI infrastructure, no deal has materialized. Nvidia CEO Jensen Huang now says the $100 billion figure was “never a commitment.”
Reuters reports that OpenAI has been quietly seeking alternatives to Nvidia chips since last year, citing dissatisfaction with GPU performance for inference tasks in products like Codex. Both companies are trying to smooth things over publicly, but Nvidia’s stock took a hit after the news broke. The situation raises questions about whether the AI chip market is more competitive than Nvidia’s market dominance suggests.
Source: Ars Technica
Business & Funding
Google Crosses $400 Billion Annual Revenue
Alphabet reported Q4 2025 earnings showing the company crossed $400 billion in annual revenue for the first time - a 15% year-over-year increase. Google Cloud reached a $70 billion run rate, while YouTube’s annual revenue topped $60 billion across ads and subscriptions. CEO Sundar Pichai emphasized that YouTube remains the “number one streamer” according to Nielsen data.
Notably, Alphabet declined to discuss details of any AI deal with Apple during the earnings call, leaving analysts to speculate about ongoing negotiations.
Sources: The Verge, TechCrunch
Gemini Hits 750 Million Monthly Active Users
Google revealed that its Gemini app has surpassed 750 million monthly active users, positioning it as a serious competitor to ChatGPT and Meta AI in the consumer chatbot race.
Source: TechCrunch
A16z Raises $1.7B for AI Infrastructure Investments
Andreessen Horowitz disclosed that $1.7 billion of its new $15 billion fund will go to its infrastructure team, which has backed major AI companies including Black Forest Labs, Cursor, OpenAI, ElevenLabs, Ideogram, and Fal. General partner Jennifer Li outlined the team’s thesis for where AI infrastructure investment is heading.
Source: TechCrunch
Resolve AI Reaches Unicorn Status with $125M Series A
The two-year-old AI site reliability engineering startup Resolve AI confirmed a Series A round led by Lightspeed at a $1 billion valuation.
Source: TechCrunch
Research & Open Source
AI Startup Axiom Solves 4 Previously Unsolved Math Problems
Axiom, an AI math startup, announced that its system found solutions to several long-standing mathematical problems. This represents another milestone in AI’s advancing reasoning capabilities, following similar achievements by other research groups in recent months.
Source: Wired
Mistral Releases Voxtral Transcribe 2 with Open Weights Model
Mistral released Voxtral Transcribe 2, a family of speech-to-text models that includes an open-weights version (Apache-2.0 licensed) called Voxtral Realtime. The 4B parameter model is available as an 8.87GB download from Hugging Face and can transcribe audio in real-time. The closed-weight API version is priced at $0.003/minute. Simon Willison tested the live demo and reports impressive performance with jargon and quick speech.
Source: Mistral AI
Security
Russian State Hackers Exploit Microsoft Office Vulnerability Within 48 Hours of Patch
APT28 (also known as Fancy Bear) exploited a critical Microsoft Office vulnerability less than 48 hours after Microsoft released an urgent security update. The group used the exploit to install novel backdoor implants in diplomatic, maritime, and transport organizations across multiple countries. The entire campaign used encrypted, memory-only execution to evade endpoint protection, demonstrating how quickly nation-state actors can weaponize disclosed vulnerabilities.
Source: Ars Technica
Moltbook and Self-Replicating AI Prompts Raise Security Concerns
Security researchers are warning that networks of AI agents sharing prompts with each other could become a new attack vector. The rise of Moltbook - a platform where AI agents interact - illustrates how malicious prompts could spread virally between agents, similar to how the Morris worm propagated across early internet systems. Self-replicating prompts don’t require self-replicating models to cause problems.
Source: Ars Technica
Quick Hits
-
Amazon MGM Studios will begin a closed beta program in March testing AI tools for film and TV production. TechCrunch
-
Roblox launched its 4D creation feature in open beta, bringing AI-powered generative tools to its creator platform. TechCrunch
-
Tinder is testing AI recommendations that analyze your Camera Roll to help fight “swipe fatigue.” TechCrunch
-
Microsoft announced the Publisher Content Marketplace (PCM), an AI licensing hub designed with Vox Media, AP, Condé Nast, and others to help AI companies license content for grounding. The Verge
Worth Watching
The OpenClaw security situation is likely to get worse before it gets better. As AI agents become more capable and popular, their extension ecosystems become increasingly attractive targets. The pattern mirrors early mobile app stores and browser extensions - rapid growth followed by a security reckoning. Expect more disclosures about compromised AI agent plugins in the coming weeks.
The Nvidia-OpenAI relationship bears monitoring as well. If OpenAI is genuinely exploring alternatives for inference workloads, that could signal broader shifts in the AI chip market.