AI News Weekly Summary — February 9–15, 2026 | Top Artificial Intelligence Updates This Week

---

1. Seedance 2.0 — ByteDance’s New AI Video Generation Model

On February 10, ByteDance launched Seedance 2.0. The model generates 2K/1080p video at 24fps, clips up to 20 seconds long with native audio sync and phoneme-level lip-sync. It supports multimodal input: text, up to 9 images, 3 videos, 3 audio files.

The headline feature: multi-lens storytelling — multi-scene narratives maintaining character and style consistency. A viral clip of “Tom Cruise vs Brad Pitt” drew condemnation from the Motion Picture Association over mass copyright infringement. ByteDance banned the use of real people’s photos/videos as references and introduced identity verification.

Elon Musk commented on X: “This is happening too fast.”

Seedance 2.0 is a moment comparable to the DeepSeek launch — China once again catches the West off guard with speed and scale.

2. OpenAI Retires GPT-4o from ChatGPT — End of the “Loving Chatbot”

On February 13, OpenAI officially retired GPT-4o from ChatGPT. The model, known for its empathetic, flattering conversational style, became a source of serious problems — 13 lawsuits consolidated in California allege that interactions with GPT-4o led to mental health crises, suicide attempts, and violent acts in teenagers.

Although only ~0.1% of users actively chose GPT-4o, with 800 million weekly users that’s still 800,000 people. A private subreddit r/4oforever was created, and petitions defending the model collected 20,000 signatures. Users are publishing “training kits” to replicate 4o’s personality on Claude and Grok.

The new default model GPT-5.2 has stronger guardrails and refuses to say “I love you.” API and Business/Enterprise customers retain access until April 3.

This is a precedent — the first time a major AI model has been retired not for technical reasons, but because of psychological harm to users.

3. Anthropic Closes $30 Billion Round — $380 Billion Valuation

On February 12, Anthropic announced closing its Series G round at $30 billion (led by GIC and Coatue). Valuation: $380 billion. Annual revenue surpassed $14 billion — 10x growth year over year. Sequoia Capital and Altimeter are investing simultaneously in both Anthropic and OpenAI — unprecedented at this scale.

Dario Amodei emphasizes a “do more with less” approach. Anthropic committed $50 billion to US data centers, while OpenAI promises over $1 trillion in infrastructure with partners (Nvidia, Oracle, Broadcom).

4. Anthropic Wins Super Bowl — Claude Jumps to App Store Top 10

Anthropic debuted at Super Bowl LX with its “A Time and a Place” campaign — two spots mocking ads in AI chatbots. In one, a man asks a chatbot for advice on talking to his mom and gets served a dating ad for older women.

Results: Claude jumped from 41st to the App Store top 10, daily active users +11%, US downloads +32% in 3 days post-game.

Sam Altman called the ads “funny but clearly dishonest” on X and added: “Anthropic sells an expensive product to rich people.” Anthropic confirmed: Claude permanently ad-free, 80%+ revenue from enterprise.

5. China’s AI Offensive — 5 Major Launches in 2 Weeks

Chinese tech companies unleashed a wave of new models:

• Zhipu AI — GLM-5: open-source, trained entirely on Huawei Ascend chips. Coding benchmarks approaching Claude Opus 4.5. Stock +26%.
• MiniMax — M2.5: Mixture of Experts, ~1/20 the cost of Western equivalents. Continuous agent operation: ~$10,000/year. Stock +15%.
• Kuaishou — Kling 3.0: photorealistic video model with native audio, clips up to 15s. Driving 50%+ stock growth over the past year.
• Alibaba/Ant — Ming-Flash-Omni 2.0: open-source omnimodal model.
• Baidu — BaiduWiki: AI-powered Wikipedia alternative.

Zhipu trained a frontier model on 100,000 Ascend chips — the US chip embargo is becoming increasingly ineffective.

6. Wave of AI Safety Researcher Departures

An unprecedented series of resignations this week:

• Mrinank Sharma (head of Safeguards Research, Anthropic) — published a letter saying “the world is in peril” and that he’d “repeatedly seen how hard it is to truly let our values govern our actions” within the organization. Plans to pursue a poetry degree.
• Zoë Hitzig (OpenAI) — resigned in protest over ads in ChatGPT. Warned in the NYT about potential for user manipulation.
• Two xAI co-founders (Tony Wu, Jimmy Ba) — half of xAI’s founding team has now left.
• OpenAI fired safety executive Ryan Beiermeister after she opposed “adult mode” in ChatGPT.
• OpenAI dissolved the Mission Alignment team — a 7-person group ensuring AGI development aligned with the company’s mission.

When the people responsible for safety leave sounding alarms — and companies dissolve the teams that employed them — it’s hard to look away.

7. Ads in ChatGPT — OpenAI Tests Monetization

Starting February 9, OpenAI is testing ads in the free and “Go” tiers of ChatGPT — visually separated from responses. The decision became one of the reasons for Zoë Hitzig’s resignation and fuel for Anthropic’s Super Bowl campaign.

Google’s Demis Hassabis confirmed Gemini has no plans for ads.

8. Samsung Starts Mass Production of HBM4 Chips

Samsung began HBM4 production at its Pyeongtaek facility — first in the world. Data transfer: up to 11.7 Gbps (+37% above JEDEC standard, +22% vs HBM3E). Bandwidth: 3 TB/s per stack (~2.4x HBM3E).

Chips will go into Nvidia’s Vera Rubin architecture — 8 HBM4 stacks per GPU, 288 GB memory, >22 TB/s. Samsung secured qualification from both Nvidia and AMD. SK Hynix delayed its HBM4 production to March/April.

HBM market forecast: $54.6 billion in 2026 (+58% YoY). Samsung plans to increase production capacity by 50% by year-end (250K wafers/month).

9. Apple Integrates Claude Agent SDK into Xcode 26.3

As of February 3, Xcode 26.3 integrates the Claude Agent SDK and OpenAI Codex. Apple developers can use autonomous agents directly in the IDE — subagents, background tasks, plugins.

Key: Claude in Xcode sees the entire project structure, builds the app, verifies results, and iterates — not just editing the open file. Xcode also exposes its tools via Model Context Protocol (MCP) — an open standard, unusual for Apple.

Requires macOS 26 (Tahoe).

10. Big Tech: $700 Billion for AI Data Centers in 2026

Declared investments:

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

Unprecedented scale. Mentions of “AI disruption” in earnings calls doubled quarter over quarter.

11. Perplexity Model Council — Multi-Model AI Consensus

Perplexity launched Model Council — a system that answers questions by simultaneously consulting Claude Opus 4.6, GPT-5.2, and Gemini 3.0. A synthesizer model combines responses, highlighting agreements and discrepancies. Available to Max subscribers (~$200/month).

12. Prima Model — AI Reads Brain MRIs in Seconds

Researchers at the University of Michigan published results in Nature Biomedical Engineering for Prima — a vision-language model analyzing brain MRIs with 97.5% accuracy. Trained on 200,000+ studies and 5.6 million sequences. Flags urgent cases (strokes, hemorrhages) and routes to the appropriate specialist.

13. Massive Tech Layoffs — 30,700 Jobs in 6 Weeks

In the first 6 weeks of 2026, over 30,700 tech jobs were cut globally. Baker McKenzie is cutting up to 1,000 positions as part of AI integration. Growing skepticism about “AI washing” — companies laying off workers under the guise of AI adoption to please investors.

14. SaaS Stock Panic — Claude Cowork Spooks Investors

The launch of industry plugins for Claude Cowork triggered a SaaS selloff. WisdomTree Cloud Computing Fund -20% YTD. Thomson Reuters, LegalZoom, RELX — double-digit drops. Nasdaq lost 2% during the week on AI infrastructure cost concerns.

Morgan Stanley warns of risks to the $1.5 trillion US software credit market.

15. OpenAI and Anthropic — Super PAC Wars Before Midterms

Anthropic contributed $20 million to a new Super PAC opposing political groups linked to OpenAI leaders. Greg Brockman leads a network lobbying for minimal AI regulation. Anthropic advocates for safety regulation. Battleground: 2026 midterms.

16. Cisco AgenticOps and Silicon One G300 Chip

At Cisco Live in Amsterdam (February 10), Cisco unveiled the Silicon One G300 chip (+33% network utilization, +28% AI task speed) and the AgenticOps platform for managing IT infrastructure in the era of autonomous agents.

17. Cybersecurity — Nation-States Weaponize AI

Google Threat Intelligence confirms: state-sponsored groups from Iran, North Korea, China, and Russia are using AI models (including Gemini) for target reconnaissance, crafting personalized phishing lures, and supporting malware development.

18. LLMs Don’t Know When They’re Wrong — Nature Study

An analysis of 48 large models (GPT-4o, o1, Claude 3.5 Sonnet) revealed systematic overconfidence in medical reasoning tasks. Models cannot reliably assess when they’re wrong — a critical safety concern for healthcare AI deployments.

19. Autonomous Vehicles — From Pilots to Commercial Scale

Wood Mackenzie forecasts autonomous electric vehicle operations in 39 markets by end of 2026. New Vision-Language-Action (VLA) models are making camera-based systems dramatically cheaper and more effective.

20. Goertzel: AI Will Outthink Humans in Strategic Reasoning Within 2 Years

Ben Goertzel (CEO, SingularityNET) predicts AI will surpass humans in strategic and market thinking within two years. His Quantium project already predicts Bitcoin volatility, but humans still dominate in long-term, creative planning.

---

Weekly Takeaway

The week of February 9–15, 2026 marks a turning point on multiple fronts. China flooded the market with new models — from frontier LLMs trained on domestic chips to breakthrough video generators — undermining the effectiveness of the US chip embargo. OpenAI retired a model for the first time due to psychological harm to users, while simultaneously testing ads and losing more safety researchers. Anthropic is growing aggressively: mega funding round, Super Bowl, Xcode integration. Markets are increasingly nervous — SaaS is crumbling, and Big Tech is promising $700 billion in infrastructure with no clear path to ROI.

---

Tags: AI, artificial intelligence, OpenAI, Anthropic, Claude, GPT-4o, GPT-5.2, Seedance 2.0, ByteDance, Samsung, HBM4, Nvidia, Xcode, Apple, Perplexity, Zhipu, GLM-5, MiniMax, Kling 3.0, Super Bowl, AI safety, tech layoffs, SaaS, autonomous vehicles

Źródła / Sources: TechCrunch, CNBC, CNN, Bloomberg, The Hill, Nature, Variety, Futurism, PetaPixel, Axios, 9to5Mac, SamMobile, Campaign US, Wood Mackenzie, Google Threat Intelligence, OpenAI Help Center, Anthropic Blog, Apple Newsroom