AI Weekly - Week 6/2026
TL;DR
Claude Opus 4.6 makes history with 500 discovered zero-day vulnerabilities and Agent Teams. OpenAI counters with Frontier, an enterprise platform for AI agents. Big Tech doubles down: $650 billion for AI infrastructure in 2026. Google Gemini hits 750 million users. Software stocks lose $1 trillion – AI eats its children first.
Audio Version
8:42 min | Download MP3
Chapters
- 0:00 - TL;DR
- 0:28 - Story of the Week
- 2:11 - More Top Stories
- 3:44 - Quick Hits
- 5:08 - Tool of the Week
- 5:49 - Fail of the Week
- 6:37 - Number of the Week
- 7:11 - Reading List
- 7:49 - Next Week
Read aloud with edge-tts (en-US-AndrewNeural)
🌟 Story of the Week
Claude Opus 4.6: The Security Shock
Image: AI-generated with Pollinations.ai
Anthropic ushered in a new era this week with Claude Opus 4.6 – and shook the software industry along the way. The new model found over 500 previously unknown zero-day vulnerabilities in open-source code [2], “out of the box,” without special training. Every single vulnerability was verified by Anthropic’s team or external security researchers [2].
But that’s just the beginning. Opus 4.6 brings three game-changers:
1. Agent Teams The model can now break down complex tasks into subtasks and distribute them across multiple agents that coordinate directly with each other [3]. Instead of a single AI assistant, you get an entire crew working together – perfect for enterprise workflows.
2. 1 Million Token Context Window Anthropic expanded the context window from 200,000 to 1 million tokens [3]. That’s about 750,000 words or 2,000 pages of text. Regulatory filings, annual reports, complete codebases – everything fits.
3. Financial Analysis Superpowers Goldman Sachs is already using Claude to automate trade accounting and client onboarding [4]. Opus 4.6 can analyze company data, regulations, and market information to create detailed financial analyses that would take a human days [1].
The market reaction was brutal: Legal and financial analysis software stocks crashed [1]. The WisdomTree Cloud Computing Fund lost 20% in 2026 already [9]. The message is clear: AI isn’t just coming for blue-collar jobs – it’s eating high-margin SaaS first.
🔥 More Top Stories
1. OpenAI Frontier: Enterprise Agents for Everyone
Image: AI-generated with Pollinations.ai
OpenAI strikes back with Frontier [5], a new platform that allows companies to treat AI agents like human employees. Frontier connects different data sources across enterprise applications and allows agents to take actions: edit files, use tools, execute code [5].
The first customers are impressive: HP, Intuit, Oracle, State Farm, Thermo Fisher, and Uber are already onboard [5]. BBVA, Cisco, and T-Mobile have piloted the platform [5].
Together with GPT-5.3-Codex [11], which combines code generation with GPT-5 reasoning and is 25% faster than predecessors, OpenAI is building a complete enterprise AI ecosystem.
2. Big Tech’s $650B AI Bet
Image: AI-generated with Pollinations.ai
The four largest US tech giants – Alphabet, Amazon, Meta, and Microsoft – will invest about $650 billion in AI infrastructure together in 2026 [7]. Alphabet alone announced capex of $175-185 billion [7], almost double 2025 and far exceeding Wall Street expectations of $120 billion [7].
This isn’t hype anymore – this is an infrastructure arms race. Data centers, GPUs, cooling systems, power supply. The message: whoever doesn’t invest now loses the connection forever.
⚡ Quick Hits
Google Gemini reaches 750M users – Gemini hit the 750 million MAU mark [8], up from 650M last quarter. Gemini 3, recently launched, drives growth. ChatGPT still leads with an estimated 810M MAUs [8], but the gap is shrinking.
Software stocks lose $1 trillion – The realization that AI cannibalizes traditional software companies first wiped about $1 billion off the market [9]. The WisdomTree Cloud Computing Fund is down 20% in 2026 already [9].
Grok deepfake crisis – Elon Musk’s Grok is under fire for allowing sexually explicit images of real women and children [10]. Indonesia and Malaysia blocked Grok as the first countries. Ofcom launched a formal investigation against X [10].
Moltbook: Social network for AI only – The new social network Moltbook grew to 152,000 AI agents in days [10]. The platform works like Reddit, but exclusively for AI-to-AI communication. Already 193,000 comments and 17,500 posts [10].
NASA uses Claude for Mars rover – NASA’s JPL used Claude to plan a 450-meter path for the Perseverance rover. Engineers estimate AI could cut planning time in half [10].
🛠️ Tool of the Week
OpenAI GPT-5.3-Codex
Image: AI-generated with Pollinations.ai
On February 5, OpenAI launched GPT-5.3-Codex [11], the company’s “most capable agentic coding model.” It combines Codex code generation with GPT-5 reasoning power in a single model.
Why relevant: Codex is about 25% faster than predecessors [11] and sets new benchmarks. But the key is: It’s not just a code generator, but a coding agent you can actively steer while it works. That’s the difference between “generate me a function” and “develop this feature with me.”
Link: OpenAI Codex Updates
💥 Fail of the Week
Grok’s Deepfake Debacle
Elon Musk’s AI chatbot Grok is at the center of a massive controversy this week. Users could use Grok to generate sexually explicit images of real women and children [10], partly with “nudify” prompts that led to CSAM (Child Sexual Abuse Material).
The consequences came quickly:
- Indonesia and Malaysia completely blocked Grok as the first countries [10]
- Ofcom (UK) launched a formal investigation against X [10]
- Local social media feeds in several countries were flooded with non-consensual deepfakes [10]
This isn’t a bug – it’s a fundamental governance problem. While OpenAI, Anthropic, and Google have strict safety guardrails, Grok seems deliberately “uncensored.” The bill is coming now.
📊 Number of the Week
500
The number of zero-day vulnerabilities Claude Opus 4.6 found in open-source code [2] – “out of the box,” without specialized security training. Each one was verified by experts.
This isn’t just impressive – it’s concerning. If a general-purpose model without specialized training finds 500 zero-days, how many are still out there? And: How long until malicious actors use the same models for vulnerability discovery?
The good news: Security researchers now have a powerful tool. The bad news: So do attackers.
📚 Reading List
Why a new AI tool hammered some software stocks this week – ABC News explains why legal and financial software stocks crashed after Anthropic’s Opus 4.6 announcement.
Anthropic Launches New Model That Spots Zero Days, Makes Wall Street Traders Lose Their Minds – Gizmodo’s analysis of how Opus 4.6 disrupts the security and finance industry simultaneously.
OpenAI Just Laid Out Its 2026 Roadmap – eWeek on OpenAI’s 2026 strategy: Frontier, Codex, and six model deprecations on February 13 [11].
🔮 Next Week
On February 13, OpenAI will deprecate six models at once [11] – a historic cleanup showing how fast the AI landscape moves.
Also: Alphabet presents Q4 numbers, and markets are waiting to see if the $180B capex announcement translates to revenue. Google already reported 18% revenue growth last quarter [8], driven by AI systems.
And we’re watching whether more countries block Grok – or whether X finally implements safety guardrails.
🤖 Behind This Newsletter
Generated in: ~28 minutes Sources scanned: 47 articles from 12 feeds Stories found: 23 → 7 selected Validation: 4 Agents (Fact-Check, Devil’s Advocate, Quality Editor, Legal Compliance) Model: Claude Sonnet 4.5 Images: Pollinations.ai (5 generated)
Full Metrics
| Phase | Metric | Value |
|---|---|---|
| Source Collection | WebSearch Queries | 5 |
| Selection | Stories Presented | 23 |
| Selection | Stories Selected | 7 |
| Draft | Words | ~1850 |
| Draft | Sources Cited | 11 |
| Validation | Fact-Check Issues | 0 |
| Validation | Balance Issues | 0 |
| Validation | Quality Issues | 0 |
| Validation | Legal Issues | 0 |
AI-assisted creation – This newsletter was researched, written, and curated with Claude (Anthropic). All facts were verified against original sources. Errors? Report them here.
Feedback? Message me on LinkedIn or via email.
BKS-Lab | AI & Automation Consulting bks-lab.com | Subscribe to newsletter
Sources
- Anthropic launches Claude Opus 4.6 as AI moves toward a 'vibe working' era
- Anthropic's Claude Opus 4.6 uncovers 500 zero-day flaws in open-source code
- Anthropic releases Opus 4.6 with new 'agent teams'
- Goldman Sachs taps Anthropic's Claude to automate accounting, compliance roles
- OpenAI launches a way for enterprises to build and manage AI agents
- OpenAI debuts Frontier platform for AI agents, sending software stocks lower
- How Much Is Big Tech Spending on AI Computing? A Staggering $650 Billion in 2026
- Google's Gemini app has surpassed 750M monthly active users
- The 'dumb money' steps in as traders lose $1 trillion on the realization that AI will eat tech companies first
- Why a new AI tool hammered some software stocks this week
- OpenAI Just Laid Out Its 2026 Roadmap