AI News Weekly Summary — February 16–22, 2026 | Top Artificial Intelligence Updates This Week

1. Pentagon vs. Anthropic — Military AI Showdown

The Pentagon is threatening to sever ties with Anthropic and designate the company a “supply chain risk” — a penalty usually reserved for foreign adversaries. The reason: Anthropic refuses to lift restrictions on mass surveillance of Americans and fully autonomous weapons.

Claude is the only AI model currently available in the US military’s classified systems. The DOD contract with Anthropic is worth $200 million. Tensions escalated after Claude was used (via Palantir) in the operation to capture Venezuelan President Nicolás Maduro.

Pentagon CTO Emil Michael publicly called Anthropic’s restrictions “undemocratic” and urged the company to “cross the Rubicon.” OpenAI, Google, and xAI have all agreed to drop their civilian guardrails for military use.

Dario Amodei responded with an essay stating that “democracies have a legitimate interest in some AI-powered military tools,” but “we should do so carefully and within limits.”

This may be the defining AI dispute of 2026 — its outcome will set norms for guardrails, audits, and operational flexibility in all future defense AI contracts.

2. Google Gemini 3.1 Pro — Doubled Reasoning Performance

On February 19, Google DeepMind released Gemini 3.1 Pro. It scored 77.1% on ARC-AGI-2 — more than double Gemini 3 Pro. Context window: 1M input tokens, 65,536 output tokens.

New: a three-tier thinking system (Low / Medium / High) — developers can modulate compute time. The model processes up to 900 images, 8.4 hours of audio, and 1 hour of video per prompt.

Gemini 3.1 Pro outscores Claude Opus 4.6, Sonnet 4.6, GPT-5.2, and GPT-5.3-Codex on the majority of benchmarks. Opus 4.6 retains the lead on Humanity’s Last Exam, SWE-Bench Verified, and τ²-bench.

Available via Gemini API, Vertex AI, GitHub Copilot, Android Studio.

The “.1” increment instead of the usual “.5” signals Google is accelerating — the model race isn’t slowing down.

3. Anthropic Claude Sonnet 4.6 — New Default Model

On February 17, Anthropic released Claude Sonnet 4.6 as the new default for Free and Pro users. Context window: 1M tokens (beta). Pricing unchanged: $3/$15 per million tokens.

Key: in internal Claude Code testing, users preferred Sonnet 4.6 over Sonnet 4.5 70% of the time — and often even over Opus 4.5 from November 2025. Computer use reaches near-human capability in tasks like spreadsheet navigation and multi-step form filling.

Customers spending $100K+ annually on Claude grew 7x YoY; $1M+ customers now exceed 500 (vs. 12 two years ago).

Anthropic shipped two models in 12 days (Opus 4.6 + Sonnet 4.6).

4. Alibaba Qwen 3.5 — Chinese Open-Weight Agentic Model

On February 16, Alibaba unveiled Qwen 3.5 — 397B parameters, sparse MoE (17B active per token), native multimodal, 256K context (open-weight) / 1M tokens (Qwen-3.5-Plus). Supports 201 languages (up from 82).

Alibaba claims the model is 60% cheaper and 8x more efficient at large workloads than its predecessor. Published benchmarks show it outperforming GPT-5.2, Claude Opus 4.5, and Gemini 3 Pro.

The cadence of frontier-competitive open-weight models from China is accelerating faster than Western labs are publicly acknowledging.

5. India AI Impact Summit — World’s Largest AI Summit

February 16–21 in New Delhi: the India AI Impact Summit — the fourth global AI summit (after Bletchley Park, Seoul, Paris), and the first hosted by a Global South nation. Delegations from 100+ countries, 20+ heads of state, 60+ ministers.

Key announcements:

• India unveiled the MANAV ethical AI vision and joined the US-led Pax Silica alliance
• Reliance Industries pledged ₹10 lakh crore (~$120B) in AI investments
• Sarvam AI launched LLMs (30B and 105B) and Kaze smartglasses (tested by PM Modi)
• Government-backed BharatGen Param2 — 17B parameters, 22 Indian languages, multimodal
• OpenAI opening two new India offices; Anthropic partnering with Infosys (Claude Center of Excellence)

AI leader predictions:

• Altman: “We might be only a couple of years from early forms of superintelligence”
• Amodei: advanced AI could lead to 25% annual GDP growth for India
• Hassabis: AGI achievable in 5 years — halving his previous timeline

Viral moment: Modi lifted Altman and Pichai’s hands — but Altman and Amodei, standing side by side, raised their fists instead of holding hands.

The event was widely criticized for chaotic organization — road closures, massive queues. IT Minister Vaishnaw apologized for “problems” on day one.

6. Google Shopping Ads in AI Mode — Ads Enter AI Conversations

Google introduced shopping ads in AI Mode — its conversational search experience with 75 million daily active users. Sponsored placements appear within AI-generated responses.

New: Direct Offers — brands can offer discounts directly within conversations. Google is testing formats in retail, travel, and other verticals.

Foundation: Universal Commerce Protocol (UCP) — launched with Shopify, Etsy, Wayfair, Target, and Walmart, enabling AI agents to discover products and complete transactions across platforms.

Google is showing what the future of ads looks like — not alongside results, but inside AI conversations. This changes the game for e-commerce.

7. Google Lyria 3 — AI Music Generation in Gemini

On February 18, Google added Lyria 3 to the Gemini app (750M MAU). The model generates 30-second tracks with lyrics and cover art, with control over style, voice, and tempo. Upload a photo or video — the AI matches music to the mood.

Limitation: the model doesn’t mimic existing artists — names are treated as stylistic inspiration. All tracks watermarked with SynthID. Available in 8 languages, globally 18+.

Lyria 3 also powers YouTube Dream Track — AI soundtracks for Shorts.

8. Amazon Ring “Familiar Faces” — Facial Recognition in Doorbells

Amazon rolled out Familiar Faces — facial recognition in Ring doorbells. A catalog of up to 50 faces, personalized alerts (e.g., “Mom at Front Door”).

Ring’s Super Bowl ad (a camera network finding lost dogs) triggered backlash — users called it “the quiet rollout of a national surveillance regime.” Senator Ed Markey wrote to Amazon’s CEO demanding the technology be discontinued. A critical TikTok hit 3 million views.

Ring also terminated its partnership with Flock Safety — a firm supplying AI cameras to police and ICE.

9. xAI — $20 Billion Round and SpaceX Acquisition

Elon Musk’s xAI closed a $20B Series E (investors: Nvidia, Cisco, QIA, Saudi Humain for $3B). A month later, SpaceX acquired xAI — xAI shares converted to SpaceX stock.

xAI operates the world’s largest AI supercomputer (Colossus I and II, >1M H100 GPUs). Grok has ~600M MAU. The company is training Grok 5.

10. AI Recommendation Poisoning — New Attack Vector

Microsoft identified a new tactic: AI recommendation poisoning. Hidden instructions in “Summarize With AI” buttons inject persistent prompts into AI assistants’ memory, biasing future recommendations toward specific companies or sites.

11. LinkedIn Abandons Traditional SEO — 60% Traffic Decline

LinkedIn reports non-brand B2B traffic dropped 60% due to AI-powered search. The company abandoned traditional SEO metrics in favor of visibility-based measurements — tracking presence in AI-generated responses (mentions, citations).

LinkedIn is the canary in the coal mine — what happened to them is coming for the rest of content marketing.

12. WordPress AI Assistant — AI Built Into the Editor

WordPress launched a built-in AI assistant — text editing, image generation, page creation, and layout modification via prompts in the site editor. Integration through @ai tags in blocks.

13. Adobe Shuts Down Animate — Pivots to AI

Adobe announced the discontinuation of Adobe Animate (2D animation software) effective March 1, 2026. The company is redirecting resources to AI products. End of the Flash/Animate era after 25+ years.

14. Google Responsible AI Report 2026

On February 18, Google published its responsible AI report — highlighting the shift from reactive tools to proactive partners capable of reasoning. The report covers agents, scientific discovery, and safety.

15. Physics-Informed Machine Learning — University of Hawaiʻi Breakthrough

Researchers at the University of Hawaiʻi published a new physics-informed machine learning algorithm in AIP Advances — AI that adheres to the laws of physics when processing data. Applications: fluid dynamics, climate modeling.

16. YouTube Conversational AI on TVs

YouTube is testing conversational AI on TV screens — viewers can ask questions about the video they’re watching, request summaries, and get recommendations without interrupting playback. Testing with Premium users in the US.

17. OpenAI–Snowflake — $200 Million Partnership

OpenAI and Snowflake announced a $200M partnership — native integration of OpenAI models into the Snowflake Cortex AI platform. Goal: enterprise data + AI in one ecosystem.

18. Delhi Declaration — 70 Countries Commit

At the India AI Impact Summit, IT Minister Vaishnaw announced that at least 70 countries would sign the Delhi Declaration on AI — a document defining principles for AI cooperation and governance.

19. MiniMax M2.1 — Expanded Multi-Language Coding

MiniMax released M2.1 — an upgraded model with expanded programming support for Rust, Java, and other languages. Optimized for complex tasks and office workflows.

20. Prediction Markets — Anthropic Dominates, Google Closing In

Prediction markets: Anthropic holds 84% probability for top 1 and top 2 in model rankings by end of February. OpenAI is shifting toward coding specialization (76% for March). Google leads in long-term forecasts — 46% for March, 54% for June.

Weekly Takeaway

The week of February 16–22, 2026 is defined by clashes. The Pentagon vs. Anthropic dispute over military AI boundaries may set norms for the entire industry. On the model front, the race continues unabated: Google answers with Gemini 3.1 Pro and record benchmarks, Anthropic counters with Sonnet 4.6, and Alibaba shows that Chinese open-weight models are catching up with the West. The India AI Impact Summit exposed both ambitions (superintelligence in a few years, 25% GDP growth from AI) and reality (organizational chaos, rivalry instead of cooperation). Google embedding ads into AI conversations may be the quiet turning point for all of e-commerce.