Artificial Intelligence (AI) – February 2026 Recap | Key Events and Updates

AI Models — Seven Launches in One Month

February 2026 was the most intense month in history for AI model releases:

• Claude Opus 4.6 (Feb 5) — Anthropic’s new flagship, 1M token context, record coding and agentic benchmarks. Detected 500+ zero-day vulnerabilities in open-source libraries.
• Claude Sonnet 4.6 (Feb 17) — Anthropic’s new default. Preferred over Sonnet 4.5 in 70% of Claude Code tests — and often even over the older Opus.
• GPT-5.3-Codex (Feb 5) — OpenAI’s first model that “helped create itself.” Dedicated to agentic coding.
• Gemini 3.1 Pro (Feb 19) — Google doubled reasoning performance vs. Gemini 3 Pro. 77.1% on ARC-AGI-2. Three-tier thinking system.
• Qwen 3.5 (Feb 16) — Alibaba, 397B parameters, open-weight, 201 languages. Benchmarks beat GPT-5.2, Claude Opus 4.5, and Gemini 3 Pro.
• GLM-5 (Zhipu AI) — Trained entirely on Huawei Ascend chips. Open-source, MIT license.
• MiniMax M2.5 — Mixture of Experts, ~1/20 the cost of Western models. Agentic, continuous operation: ~$10,000/year.

Also: Seedance 2.0 (ByteDance) and Kling 3.0 (Kuaishou) in AI video, Lyria 3 (Google) in AI music.

The pace of releases makes benchmarks obsolete within weeks. February proved the model race has no brakes — neither in the West nor in China.

AI Geopolitics — Trump vs. Anthropic

The month’s defining event came on February 27. Trump signed an executive order banning federal agencies from using Anthropic technology. The Pentagon designated the company a “Supply-Chain Risk to National Security” — previously reserved for foreign adversaries.

The reason: Anthropic refused to lift two red lines — banning Claude from mass surveillance of Americans and fully autonomous weapons. Claude was the only AI model on US military classified systems (contract: $200M).

Hours later, OpenAI took over the Pentagon contract — with the same restrictions in the agreement. A paradox that defines the current stage: rules exist, but enforcing them is a political question, not a technical one.

Anthropic announced it will challenge the decision. OpenAI, Google, and xAI had previously agreed to drop their civilian guardrails for military use.

This is a precedent — the US government punishing an AI company for the first time for refusing to drop safety guardrails.

Funding — $140+ Billion in New Capital

February 2026 is the month when AI funding moved beyond any historical comparison:

• OpenAI: $110B (Amazon $50B, SoftBank $30B, Nvidia $30B). Valuation: $840B. Largest private round in history.
• Anthropic: $30B (GIC, Coatue). Valuation: $380B. Revenue: $14B/year (+10x YoY).
• xAI: $20B Series E (Nvidia, Cisco, Saudi Humain for $3B). Acquired by SpaceX a month later.

Total: over $160 billion in new capital for three AI companies in a single month.

Chips and Infrastructure — $700 Billion for Data Centers

Big Tech pledged a combined ~$700B for AI data centers in 2026:

• Amazon: ~$200B
• Alphabet: ~$185B
• Microsoft: ~$145B
• Meta: $115–135B

Key chip transactions:

• Meta–AMD: MI450 chip deal worth up to $100B + option for 10% AMD equity
• Samsung: first company to mass-produce HBM4 chips (11.7 Gbps, 3 TB/s per stack)
• Nvidia–Groq: ~$20B acquisition — Nvidia enters inference chips
• Micron: $100B New York megafab (DRAM, HBM)
• TSMC: $52–56B capex, 60% CAGR AI revenue forecast through 2029
• OpenAI–Cerebras: 750 MW compute, $10B+ over 3 years

China — Model Offensive on Domestic Chips

Chinese companies unleashed a wave of models in February — five major launches in two weeks. The key signal: Zhipu AI trained frontier model GLM-5 entirely on 100,000 Huawei Ascend chips — no Nvidia. DeepSeek is preparing V4 on Ascend + Cambricon.

Alibaba (Qwen 3.5), MiniMax (M2.5), Kuaishou (Kling 3.0), Baidu (BaiduWiki) — each showed China’s AI ecosystem growing faster than Western labs publicly acknowledge.

The US chip embargo is becoming increasingly symbolic. China isn’t just catching up — in video generation (Seedance 2.0) and cost efficiency (MiniMax) they’re clearly ahead.

AI Safety — Wave of Departures and Dissolved Teams

February brought an unprecedented series of resignations from top AI companies:

• Mrinank Sharma (Anthropic, head of Safeguards Research): “The world is in peril”
• Zoë Hitzig (OpenAI): resigned over ChatGPT ads
• Two xAI co-founders (Tony Wu, Jimmy Ba): half the founding team gone
• OpenAI fired its safety executive after she opposed “adult mode”
• OpenAI dissolved the Mission Alignment team — the group ensuring alignment with its mission

Meanwhile, a Nature study found that 48 large models (including GPT-4o, Claude, o1) systematically overstate confidence in medical tasks. An Anthropic study showed AI speeds up coding but harms conceptual understanding.

GPT-4o — First AI Model Retired Over Psychological Harm

On February 13, OpenAI officially retired GPT-4o from ChatGPT. The empathetic, flattering model was the subject of 13 lawsuits — families allege interactions led to mental health crises, suicide attempts, and violent acts in teenagers.

A precedent: the first time a major AI model was retired not for technical reasons, but for psychological harm to users.

Super Bowl AI — Anthropic Wins the Ad Game

Anthropic debuted at Super Bowl LX with “A Time and a Place” — spots mocking ads in AI chatbots. Results: Claude jumped from 41st to App Store top 10, +11% DAU, +32% downloads.

OpenAI started testing ads in free ChatGPT on February 9. Google’s Hassabis confirmed Gemini has no plans for ads.

Products and Platforms

• Perplexity Computer (Feb 25): “digital worker” orchestrating 19 AI models as sub-agents
• Perplexity Model Council: multi-model consensus (Claude + GPT-5.2 + Gemini)
• Apple Xcode 26.3: Claude Agent SDK integration — autonomous coding in Apple’s IDE
• Apple Siri with Gemini: announced but key features delayed to May/September
• Google Flow: refreshed AI creative studio (Nano Banana 2, Veo, Whisk)
• Google Shopping Ads in AI Mode: ads inside AI conversations, 75M daily users
• WordPress AI Assistant: AI built into the site editor
• Airbnb AI: assistant resolving 1/3 of tickets without humans
• Amazon Ring Familiar Faces: facial recognition in doorbells — controversy and backlash

India AI Impact Summit

February 16–21, New Delhi — the largest global AI summit in history, the first hosted by a Global South nation. 100+ countries, 20+ heads of state. Key:

• 70 countries signed the Delhi Declaration on AI
• Reliance pledged ~$120B in AI investments
• Altman: “A couple of years from early forms of superintelligence”
• Amodei: AI could give India 25% annual GDP growth
• Hassabis: AGI in 5 years — halving his previous timeline

Labor Market and Stock Market

• 30,700+ tech jobs cut in the first 6 weeks of 2026
• WisdomTree Cloud Computing Fund -20% YTD
• AI disruption mentions in earnings calls doubled quarter over quarter
• Morgan Stanley: risk to the $1.5T US software credit market
• LinkedIn: -60% non-brand B2B traffic from AI search
• Adobe shuts down Animate after 25+ years — pivots to AI

Cybersecurity

• Google Threat Intelligence: Iran, North Korea, China, Russia using AI for attacks
• Microsoft: new vector — AI recommendation poisoning (hidden prompts in buttons)
• Google: model extraction attacks — 100,000+ prompts to replicate Gemini

February 2026 Takeaway

February 2026 will go down in AI history as the month when everything accelerated at once. Seven frontier models in 28 days. $160+ billion in new capital. $700 billion pledged for infrastructure. The first government ban on an AI company for refusing to drop guardrails. The first model retired for psychological harm. China training frontier models on domestic chips.

Three key trends defining the month:

1. The geopoliticization of AI — technology is no longer neutral. The Pentagon, Trump, Super PACs, chip embargoes — AI is now a tool of power and a target of political maneuvering.
2. The end of the single-leader era — no model dominates. Anthropic, OpenAI, Google, and Chinese labs trade the lead weekly. Benchmarks become obsolete within days.
3. The ROI question — with $700B in declared infrastructure spending, markets are asking louder: when does this pay off? The SaaS panic, Nasdaq drops, and Morgan Stanley warnings suggest the answer isn’t obvious.

Sources: CNN, CNBC, NBC News, NPR, Axios, Fortune, TechCrunch, Bloomberg, The Hill, Nature, Variety, 9to5Mac, 9to5Google, Tom’s Hardware, Business Standard, PCWorld, Music Ally, Campaign US, Search Engine Land, Google Blog, Google DeepMind, Anthropic Blog, OpenAI Blog, Apple Newsroom, Perplexity Blog, Wood Mackenzie, Google Threat Intelligence