Anthropic officially confirmed it will take the U.S. Government to court following an unprecedented “national security” blacklist.

The legal battle stems from a dramatic weekend standoff where the Trump administration designated the AI firm as a “supply chain risk,” a label typically reserved for foreign adversaries like Huawei.

The Conflict: A Standoff Over “Red Lines”

The dispute erupted after Anthropic CEO Dario Amodei refused to meet a Friday deadline (February 27) set by Defense Secretary Pete Hegseth.

The “Supply Chain Risk” Designation

After the deadline passed, the administration issued a far-reaching order:

• Immediate Federal Ban: President Trump directed every federal agency to “immediately cease all use” of Anthropic’s technology.
• Contractor Blacklist: The Pentagon barred any military contractor (including giants like Boeing and Lockheed Martin) from conducting commercial activity with Anthropic.
• The “Scarlet Letter”: By labeling a U.S. company a “supply chain risk,” the government has essentially blocked Anthropic from the entire defense ecosystem.

Anthropic’s Legal Challenge

In a defiant blog post and subsequent interviews, Anthropic called the move “legally unsound” and “retaliatory.” Their lawsuit is expected to focus on three key arguments:

1. Overreach of Authority: Anthropic argues that the Secretary of Defense lacks the statutory authority to block a U.S. company from serving other private contractors for non-military work.
2. First Amendment Violation: Amodei stated that “disagreeing with the government is the most American thing in the world,” framing the “woke AI” label as a violation of the company’s right to set its own ethical standards.
3. Misuse of the Statute: Legal experts suggest the supply chain law was never intended to be used as a “punishment” for a domestic contract dispute over safety settings.

The “Claude vs. OpenAI” Split

While Anthropic dug in, OpenAI took a different path. Hours after the ban, CEO Sam Altman announced that OpenAI had reached its own deal with the Pentagon to deploy models on classified networks. Altman claimed the deal includes safety guardrails, though the Pentagon has praised OpenAI for being “cooperative” compared to Anthropic’s “arrogance.”

“No amount of intimidation or punishment from the Department of War will change our position on mass domestic surveillance or fully autonomous weapons.” — Anthropic Official Statement

Public Backlash: The “Streisand Effect”

Ironically, the ban has triggered a massive surge in public support for Anthropic. As of today, Claude has hit No. 1 on the U.S. App Store, overtaking ChatGPT. Additionally, over 700 employees from Google and OpenAI have signed an open letter backing Anthropic’s stance against unrestricted military AI.

Perplexity AI released its first family of open-source embedding models, called pplx-embed.

Designed for web-scale retrieval and RAG (Retrieval-Augmented Generation) systems, these models aim to match the quality of industry leaders like Google and Alibaba while being significantly more efficient in terms of memory and storage.

The Two New Models

Perplexity released two specialized versions, each available in two different sizes (0.6B and 4B parameters):

1. pplx-embed-v1: Optimized for standard dense text retrieval. This is best for independent texts, single sentences, and search queries.
2. pplx-embed-context-v1: Optimized for context-aware retrieval. It embeds document chunks while considering the surrounding document context, which is particularly useful for disambiguating complex information in RAG systems.

Key Technical Innovations

The pplx-embed family introduces several features that distinguish it from traditional embedding models:

• Bidirectional Context: Unlike many models that process text only from left to right, these models are converted from Alibaba’s Qwen3 architecture into bidirectional encoders. This allows them to “see” context in both directions, improving retrieval accuracy by roughly 1%.
• Instruction-Free: Most modern embedding models require “instruction prefixes” (e.g., “Represent this query for retrieval”). Perplexity’s models require no prefixes, which prevents potential search quality degradation if prefixes aren’t used consistently between indexing and query time.
• Native Quantization: The models are trained to produce INT8 and binary embeddings natively.
  • INT8: Reduces storage requirements by 4x with no performance loss.
  • Binary: Reduces storage by 32x, making it feasible to store billions of vectors in mobile or edge applications.

Performance & Availability

Perplexity claims these models outperform major competitors on internal benchmarks that reflect “real-world” noisy web data rather than just academic sets.

The Strategy

By open-sourcing these models, Perplexity is positioning itself as an infrastructure provider rather than just a chatbot. This allows developers to build search engines using the same foundational “math” that powers Perplexity’s own multi-billion page index.

On Tuesday, February 24, 2026, a provocative research paper from Google DeepMind titled “Saving Skills in the Age of AI” proposed that AI systems should occasionally assign humans “intentional busywork.”

The proposal aims to solve the “Automation Paradox”—the risk that humans will lose the core skills and intuition required to intervene during critical AI failures because machines have taken over all routine practice.

The Strategy: “Forced Engagement”

Rather than maximizing for pure efficiency, the DeepMind researchers suggest that AI agents should optimize for a composite reward function that balances task speed with human skill maintenance.

• Deliberate Inefficiency: The AI would periodically “hand back” tasks to humans that it could easily solve itself.
• Targeted Frequency: Simulations suggest that making just 5–10% of interactions manual can significantly prevent “skill atrophy” without tanking overall productivity.
• Skill Thresholds: The AI would monitor a user’s performance and only trigger “busywork” when it detects that the human’s proficiency in a specific area is beginning to slide.

Real-World Examples from the Paper

The researchers provided several scenarios where this “busywork protocol” could be applied:

The “Moral Crumple Zone”

One of the most striking parts of the paper is the warning about “moral crumple zones.” This refers to a scenario where humans sit in a “delegation chain” simply to absorb legal liability when things go wrong, even if they no longer have the actual skills or control to prevent the error. By assigning busywork, DeepMind argues we can move from being “passive observers” back to being “competent partners.”

Reception and Criticism

The proposal has sparked a polarized debate among tech leaders:

• The Efficiency Camp: Critics, including several venture capitalists, argue that “planned inefficiency” is a regressive tax on productivity that will only slow down innovation.
• The Safety Camp: Supporters argue that as we move toward Artificial General Intelligence (AGI), the “deskilling” of the human race is a genuine existential threat that requires immediate structural safeguards.

NVIDIA is developing a “top-secret” dedicated inference computing platform designed to address the massive speed and cost demands of the next generation of AI agents.

The new system, which is expected to be officially unveiled at NVIDIA’s GTC conference in San Jose later this month, represents a major strategic shift from training-focused GPUs to specialized inference hardware.

The Strategy: Integrating Groq Technology

In a move to block competitors like Cerebras and SambaNova, NVIDIA has reportedly struck a $20 billion licensing deal with the chip startup Groq.

• Hybrid Architecture: The new platform will reportedly combine NVIDIA’s Blackwell/Rubin GPUs with Groq’s Language Processing Units (LPUs).
• Deterministic Performance: Unlike traditional GPUs that excel at parallel training, Groq’s LPU technology is designed for ultra-low latency, making it ideal for the “decoding” stage of LLMs (where words are generated one by one).
• Talent Acquisition: As part of the deal, NVIDIA has hired Groq’s founding CEO, Jonathan Ross (a key designer of Google’s original TPU), and its President, Sunny Madra.

OpenAI as the Anchor Customer

The platform is being built in close collaboration with OpenAI, which has reportedly been unsatisfied with current hardware speeds for complex tasks like software development.

• Codex Integration: OpenAI plans to use the new chip to power its Codex programming tools to compete with Anthropic’s Claude Code.
• Dedicated Capacity: This deal follows NVIDIA’s recent $30 billion investment into OpenAI, reinforcing a “cycle” where OpenAI uses NVIDIA’s capital to buy its high-end specialized chips.

Projected Impact on “Agentic AI”

NVIDIA’s pivot to a dedicated inference platform addresses the growing “Inference Gap” in the industry:

The Rubin Platform (January 2026 Announcement)

This new “secret” chip is likely a core component of the NVIDIA Rubin platform, which was teased at CES 2026. The Rubin architecture introduces:

• NVIDIA Vera CPU: A new high-performance processor.
• Inference Context Memory Storage: A new class of AI-native storage to help agents “remember” long-running tasks without re-processing entire datasets.

On Saturday, February 28, 2026, during its inaugural product showcase event called Groww Next, Bengaluru-based wealthtech giant Groww unveiled GR1, its first native AI-powered investing assistant.

Billed as an “AI co-pilot,” GR1 is designed to act as a personal research analyst that understands your specific financial context rather than just giving generic market advice.

Key Capabilities of GR1

Unlike standard LLMs, GR1 is built on an architecture that Groww calls “Responsible Intelligence,” meaning it is deeply integrated with your actual portfolio data.

• Portfolio Intelligence: It can analyze your current holdings to answer complex questions like, “Am I overexposed to the IT sector?” or “How did my portfolio specifically react to yesterday’s market dip?”
• News Sentiment Analysis: The tool scans live news and financial reports to explain why a particular stock in your watchlist or portfolio is moving.
• Personalized Insights: It generates investment ideas and “intelligent nudges” based on your individual risk profile and historical investing behavior.
• Advisory, Not Autonomous: GR1 is strictly assistive. It cannot execute trades on its own. Every action suggested by the AI requires explicit user consent through multiple “guardrail” layers.

Launch Details & Access

• Status: Currently in Beta.
• User Base: It is being tested with a limited group of 30,000 to 50,000 users before a wider rollout expected in the next 3–6 months.
• Availability: The feature is opt-in only. If you prefer a traditional, self-directed experience, the core Groww interface remains unchanged.

The “Multiple Apps” Strategy

The launch of GR1 is part of a broader 2026 strategy where Groww is moving away from the “Super App” model and launching specialized platforms:

Market Context

This move follows Groww’s massive success in 2025, where it solidified its position as India’s largest broker by active users. With over 1.6 crore customers and a Q3 FY26 net profit of ₹547 crore, Groww is using GR1 to deepen engagement with younger investors (average age 27) who prefer tech-first research over traditional relationship managers.

Anthropic officially launched a new “Import Memory” feature for Claude, specifically designed to help users migrate from other AI assistants like ChatGPT and Gemini without losing their accumulated context.

The move is being seen as a tactical play to capture users during the viral “Cancel ChatGPT” trend, which was triggered by OpenAI’s recent controversial partnership with the U.S. government for military AI applications.

How the Import Works

Anthropic has simplified the migration into a “copy-paste” process rather than a complex file export. You can find the new tool at claude.com/import-memory.

1. Extract: Claude provides a “Digital Passport” prompt. You paste this into your existing chatbot (ChatGPT, Gemini, etc.).
2. Export: The old chatbot then spits out a clean code block containing your stored preferences, bio, and work style.
3. Import: You paste that block into Claude’s Settings > Capabilities > Memory, and Claude “learns” who you are within 24 hours.

What Claude Can “Learn” From Others

The import is designed to preserve the “digital soul” or tacit understanding you’ve built with other models:

• Response Style: Instructions like “always keep it concise” or “never use emojis.”
• Professional Context: Your job title, current projects, and preferred coding languages.
• Personal Details: Your location, interests, and family context (though Claude focuses primarily on work-related memories for efficiency).

Key Details & Limitations

The Competitive Context

Anthropic has “beaten Google to the punch” with this launch. While Google has been testing a similar feature for Gemini since February 2026, it focuses on importing full chat histories, whereas Anthropic focuses on the distilled “memory” edits, which is more privacy-centric and faster to process.

Grok became the center of a viral firestorm after it “correctly” predicted the exact date of the coordinated US-Israeli strikes on Iran.

While Elon Musk celebrated the hit as the “ultimate measure of intelligence,” the reality behind the prediction is a fascinating mix of probability and a high-stakes “stress test” by the media.

The AI “War-Game” Exercise

On February 25, three days before the strikes, The Jerusalem Post conducted an experiment by asking four major AI models (Claude, Gemini, ChatGPT, and Grok) to pick a single, specific day for a hypothetical US strike on Iran.

• Grok’s Logic: Grok specifically linked its February 28 prediction to the outcome of the Geneva diplomatic talks and the 10-to-15-day deadline publicly stated by President Trump on February 19.
• The “Win”: When the strikes (Operations “Roaring Lion” and “Epic Fury”) actually began on the morning of February 28, Grok was the only model to have pinpointed that exact day.

Foresight or Lucky Guess?

Military experts and the Jerusalem Post staff have offered a grounded perspective on why Grok “won”:

1. Open-Source Intelligence (OSINT): Grok did not have access to classified Pentagon plans. Instead, it synthesized publicly available data: the buildup of US carrier groups, the breakdown of the Geneva talks, and the specific rhetoric coming from the White House.
2. Probabilistic Window: In a high-tension cycle, the “window” for a strike is mathematically small (usually a weekend or a specific diplomatic deadline). By forcing the AI to pick one day, the likelihood of one model hitting the mark was relatively high.
3. The “X” Advantage: Because Grok is integrated with X (formerly Twitter), it has a “shorter” delay in processing real-time news and public sentiment compared to other models, which may have influenced its focus on the immediate Feb 28 deadline.

Elon Musk’s Reaction

Elon Musk leaned heavily into the viral moment, sharing a link to the Jerusalem Post article with the caption:

“Prediction of the future is the best measure of intelligence.”

He also noted that February 28 was the “biggest day on X in history,” as users flocked to the platform for real-time updates on the strikes and to discuss Grok’s apparent “forecasting” ability.

The “Palliative” Paradox

Ironically, this news comes just as other reports suggest the US military actually used Claude (which “failed” the prediction test) to execute the strikes, despite Trump’s federal ban on the model. This highlights a strange divide: Grok is winning the public PR war for its “intelligence,” while Claude is being used behind the scenes for its “operational” reliability.