Top Stories
Anthropic’s Claude Code Security Tanks Cybersecurity Stocks
Anthropic unveiled Claude Code Security on February 20, and the cybersecurity sector immediately lost billions in market value. CrowdStrike dropped 8%. Cloudflare fell 8.1%. SailPoint shed 9.4%. Investors are betting that AI-powered security scanning will commoditize a significant chunk of what these companies do.
The tool works differently from traditional static analysis. Instead of pattern matching against known vulnerability signatures, Claude Code Security “reasons about your code the way a human security researcher would” - mapping component interactions, tracing data flows, and identifying weak points that rule-based scanners miss. In testing on open-source software, it found vulnerabilities that had gone undetected for decades.
The product launches as a limited research preview for Enterprise and Team customers, with expedited access for open-source maintainers. Anthropic’s pitch: attackers will use AI to find exploits faster than ever, so defenders need to move first.
The market reaction reveals a larger anxiety. If AI can perform security audits, penetration testing, and vulnerability assessment - tasks that currently require expensive human expertise - the business model of the entire security software industry comes under pressure. These companies sell tools; Anthropic is selling an analyst.
Sources: Bloomberg, SiliconANGLE, Fortune, Anthropic
Google Drops Gemini 3.1 Pro With 2x Reasoning Performance
Google released Gemini 3.1 Pro on February 19-20 - the first “.1” increment in Gemini’s release history. The company positioned it as a reasoning upgrade: on ARC-AGI-2, the model scored 77.1%, more than double Gemini 3 Pro’s performance.
The model brings Deep Think’s “upgraded core intelligence” to a broader user base. Google describes it as designed for “tasks where a simple answer isn’t enough” - synthesizing complex data, explaining multi-step processes, handling “ambitious agentic workflows.”
Availability is rolling out to the Gemini app, NotebookLM (for Pro and Ultra subscribers), and developer platforms including Google AI Studio, Vertex AI, and Android Studio. It’s launching in preview while Google validates the improvements before general availability.
The timing matters. The AI model race has accelerated to monthly releases. Google’s .1 naming signals they’re moving to more frequent updates - smaller capability jumps, faster iteration. OpenAI, Anthropic, and Chinese labs are all pushing new models every few weeks. Users benefit from rapid improvement; developers struggle to keep up with API changes.
Sources: 9to5Google, Google Blog
Apple’s AI-Powered Siri Delayed Again, Features Spread Across iOS 26.5 and 27
Apple’s long-promised Siri upgrade has hit testing problems, pushing features originally planned for iOS 26.4 (March) into iOS 26.5 (May) and iOS 27 (September). The context-aware assistant that Apple previewed - capable of “on-screen awareness” and cross-app intelligence - won’t arrive as a single update.
The new Siri uses Google’s Gemini models under a multi-year deal reportedly costing Apple $1 billion annually. Demonstrations showed Siri extracting flight information from Mail and pulling reservation details from Messages. The vision was a seamless experience where the assistant understands your screen and your data across apps.
But internal testing revealed integration issues. Apple is now spreading the capabilities across multiple releases rather than delivering them all at once. The company confirmed to CNBC that the smarter Siri is still coming in 2026 - just with uncertainty about exactly when.
This matters for Apple’s AI narrative. The company positioned itself as doing AI “the Apple way” - slower but more polished. Repeated delays undermine that pitch. Meanwhile, ChatGPT’s mobile app and Claude’s iOS integration keep improving. Every month of delay is a month where users get accustomed to non-Apple AI assistants on their iPhones.
Quick Hits
-
Sarvam AI launches India’s sovereign models: The Bangalore startup unveiled 30B and 105B-parameter models at the India AI Impact Summit, trained on Indian languages and optimized for local deployment. The 105B model “Indus” handles 22 Indian languages with voice commands. Available on iOS, Android, and web. India is racing to build AI infrastructure that doesn’t depend on US or Chinese companies. TechCrunch, Business Standard
-
OpenAI disbands mission alignment team: The seven-person team focused on ensuring AGI “benefits all of humanity” was dissolved, with members transferred elsewhere. Leader Joshua Achiam becomes OpenAI’s “chief futurist.” This is the second alignment team OpenAI has disbanded in 16 months. On the other hand, OpenAI contributed £5.6 million to the UK’s new Alignment Project. Interpret those signals as you will. TechCrunch, Resultsense
-
AI layoffs surge past 30,000 in 2026: Tech layoffs hit 30,000+ for 2026 so far, with companies increasingly citing AI as the reason. Baker McKenzie is cutting 600-1,000 support staff. Workday announced 400 cuts. Meta is shedding 1,500 from Reality Labs. But economists warn many “AI layoffs” are actually AI-washing - companies using AI as convenient cover for cuts driven by overhiring or economic pressure. Above the Law, Gulf News
-
Open source models closing the gap: GLM-5 and Kimi K2.5 now score within single-digit Quality Index points of proprietary models. Kimi K2.5 hit 96% on AIME 2025, beating most commercial alternatives on math. Chinese open-source downloads on Hugging Face have surpassed US models. The “open source is always behind” assumption no longer holds. Medium
-
78 state AI bills and counting: Six weeks into 2026’s legislative session, 27 states have introduced 78 chatbot-related bills. Most focus on child safety - prohibiting suicide encouragement, self-harm content, and grooming. Virginia and Washington passed bills out of their Senates. The Trump administration’s executive order threatens to preempt state AI laws deemed “burdensome” - but the March deadline for identifying targets is approaching with no clear action yet. Transparency Coalition
Worth Watching
The Claude Code Security market reaction deserves close attention. An 8-9% single-day drop across major cybersecurity stocks on a product announcement - not a competitor’s earnings beat - suggests investors see something structural. If AI can do security research, what else in the software industry becomes commoditized? The SaaSpocalypse thesis from early February gains more evidence.
Apple’s Siri troubles highlight a deeper problem: integrating AI assistants deeply into operating systems is hard. Google and Microsoft have the advantage of building both the AI models and the OS. Apple is licensing Gemini, adding a coordination layer between teams at different companies with different priorities. Every missed deadline reinforces that building AI-native experiences is different from adding AI features to existing products.
The alignment team dissolution at OpenAI creates a pattern. Two teams disbanded in 16 months, both focused on long-term AI safety. The company is simultaneously contributing to external alignment research (UK’s Alignment Project) while eliminating internal oversight. That’s a shift from “we ensure our own AI is safe” to “we fund others to study AI safety.” Whether that’s pragmatic resource allocation or a change in priorities, the governance implications matter.