OpenAI vs DeepSeek, Orchid hack & more (February 13, 2026)

Good morning,

Geopolitics is heating up in the model race, capital continues pouring into frontier labs, and AI agents are moving deeper into real-world systems. But one investigation this week shows the risks when convenience outruns security.

Let’s dive in 👇

━━━

Powered by Aloa • AI Research & Education Hub

━━━

⚔️ Model Wars & Capital Escalation

🧨 OpenAI accuses DeepSeek of distillation

OpenAI accuses DeepSeek of distilling US models to gain a competitive edge, escalating tensions between American and Chinese AI labs. The dispute centers on whether outputs from advanced US systems were used to train rival models, raising intellectual property and enforcement concerns. If validated, this could further intensify export controls and geopolitical pressure around frontier AI development.

💰 Anthropic raises another $30B

Anthropic raises another $30 billion in Series G, pushing its valuation to $380 billion and reinforcing investor conviction in large-scale AI labs. The funding gives Anthropic significant flexibility to expand compute infrastructure, research capacity, and enterprise partnerships. Capital concentration at this scale signals continued belief that frontier models will dominate long-term value creation.

🧠 Codex gets a dedicated chip

A new version of OpenAI’s Codex is powered by Cerebras, a purpose-built chip optimized specifically for code generation workloads. The move reflects tighter vertical integration between model design and hardware performance. Custom silicon may become a structural advantage as AI companies optimize for speed, efficiency, and developer productivity.

🔐 Agents, Upgrades & Security Reality

⚠️ Orchid vibe-coding platform exposed security risks

Orchids, a popular “vibe-coding” platform for non-technical users, was shown to contain a flaw enabling a zero-click system takeover during a BBC test. Researcher Etizaz Mohsin inserted hidden code into a live project, gaining unauthorized access to the reporter’s machine. Experts warn that giving AI coding agents deep system access creates serious new security risks as adoption accelerates.

🔍 Gemini 3 “Deep Think” upgrade

Gemini 3 Deep Think upgrade introduces stronger multi-step reasoning and deeper contextual processing across longer tasks. The update emphasizes sustained chains of thought and improved logical consistency in complex workflows. This positions Gemini more directly against other reasoning-focused frontier systems competing for enterprise trust.

🏭 Didero raises $30M for agentic procurement

Didero lands $30M to put manufacturing procurement on autonomous rails using AI agents to manage supplier workflows and sourcing decisions. The company targets industrial bottlenecks where manual coordination slows execution and increases costs. Investors continue backing agentic systems that embed directly into operational infrastructure rather than experimental use cases.

🛠️ Tools of the Day

→ Lovon – AI voice generation for scalable audio content
→ Zenmux – AI-native collaboration workspace
→ Arena – Browser built for AI research and automation

⚡ Quick Hits

→ Spotify says its best developers stopped manual coding
→ AI video of Tom Cruise sparks confusion
→ Legora in talks to triple valuation
→ Didero lands $30M for procurement automation

🧾 TLDR

OpenAI’s dispute with DeepSeek adds geopolitical tension to the model race while Anthropic’s $30B raise underscores sustained capital concentration in frontier labs. Codex moving onto custom silicon signals deeper stack control across hardware and software. Meanwhile, the Orchids zero-click exploit highlights how agentic convenience may be scaling faster than security discipline.

Cheers,
David