Presented by
Good morning, AI enthusiasts. Imagine hiring 24,000 fake accounts to extract a competitor's AI capabilities one carefully crafted prompt at a time — that's exactly what Anthropic says three Chinese labs did, generating over 16 million exchanges to steal what they couldn't build on their own.
In today's recap:
Three Chinese labs exposed stealing Claude capabilities
Anthropic's security scanner sends cyber stocks sliding
Run parallel media tasks with Claude Cowork
OpenAI brings consulting giants into enterprise AI
4 new AI tools, prompts, and more
ANTHROPIC
Three Chinese AI labs caught stealing from Claude
Getty Images
Recaply: Anthropic just exposed three Chinese AI companies running an industrial-scale scheme to steal Claude's capabilities, with DeepSeek, Moonshot, and MiniMax generating over 16 million exchanges through 24,000 fraudulent accounts.
Key details:
Labs used a technique called distillation, generating large volumes of crafted prompts to extract specific capabilities from Claude, with outputs collected for direct model training or reinforcement learning at scale.
MiniMax led the effort with over 13 million exchanges, followed by Moonshot AI at 3.4 million and DeepSeek at 150,000, each targeting Claude's agentic reasoning, tool use, and coding capabilities.
Anthropic detected MiniMax's campaign while still active. When a new Claude model launched mid-campaign, MiniMax pivoted within 24 hours, redirecting nearly half its traffic to the latest system, according to Anthropic's report.
Anthropic published its findings on February 23, 2026, and is now sharing technical indicators with other AI labs, cloud providers, and authorities while tightening verification for commonly exploited account types.
Why it matters: The gap between export controls and reality has driven AI policy debates for years. But Anthropic's report puts hard numbers on what was mostly theory: Chinese labs aren't just innovating independently, they're systematically extracting capabilities from American models at industrial scale. That reframes the DeepSeek story significantly. If distilled models strip out safety guardrails along with the capabilities, the national security implications extend well beyond competitive advantage into genuinely dangerous territory.
PRESENTED BY BELAY
Speed Doesn’t Replace Strategy.
AI can surface the numbers in seconds, but numbers alone don’t create clarity.
In fact, many leaders have more financial data than ever yet less clarity about what to do next.
The real challenge isn’t reporting. It’s interpretation. Context. Judgment.
BELAY created the free guide The Future of Financial Leadership to explore why automation is a tool — not a replacement — for experienced financial oversight.
Inside, you’ll learn how the right human support brings structure to your numbers, confidence to your decisions, and focus to your growth strategy.
At BELAY, our U.S.-based Financial Experts help leaders move beyond dashboards and into decisive action.
Because insight doesn’t drive a business forward. Leadership does.
ANTHROPIC
Claude Code Security sends cybersecurity stocks sliding
Recaply: Anthropic just launched Claude Code Security in a limited research preview, a codebase scanner that found 500+ vulnerabilities in open-source projects and sent cybersecurity stocks tumbling by up to 11%.
Key details:
Claude Code Security reasons through code the way a human security researcher would, tracing how data moves through applications to catch complex vulnerabilities that rule-based static analysis tools typically miss, with each finding verified before reaching analysts.
CrowdStrike, Datadog, and Zscaler each fell around 11% on Monday, while Okta and Fortinet dropped roughly 6%; Anthropic's security team found the 500+ vulnerabilities using Claude Opus 4.6 in production open-source codebases.
Robert W. Baird analyst Shrenik Kothari called the reaction "panic-driven" and "narrative-led," noting Claude Code Security doesn't handle real-time tasks like detecting live intrusions or stopping attacks in progress.
Claude Code Security launched February 20 and is now open to Enterprise and Team customers; open-source maintainers can apply for free, expedited access at claude.com.
Why it matters: While Claude Code Security focuses narrowly on pre-deployment code review, the market reaction shows how much investors fear AI expanding into traditional security territory. Robert W. Baird analyst Shrenik Kothari called it a "panic-driven, narrative-led selloff," noting the tool doesn't handle real-time threats. The distinction is real for now. But as AI gets better at finding vulnerabilities before they're exploited, the gap between pre-deployment tools and live security platforms may shrink faster than analysts expect.
TUTORIAL
Run parallel media tasks with Claude Cowork
Recaply: In this tutorial, you will learn how to use Claude Cowork to process images and videos in parallel, compressing files and extracting audio simultaneously without managing each task step by step.
Step-by-step:
Update your Claude Desktop app to the latest version and click the Cowork tab to get started — available on paid plans (Pro at $20/month or Max at $100/month for heavier workloads).
Create a new folder with the images and videos you want to process, then in Cowork click "Work in a folder" and select it to give Claude direct access to your local files.
Type your first task prompt: "Save copies of the images in this folder, then reduce the file size of each by at least 50%" — Claude will start working and show real-time progress in the task view.
Before the first task finishes, submit a second prompt: "Extract the audio from the video in this folder and save it as an MP3" — Cowork coordinates both tasks in parallel using sub-agents.
Once complete, open your folder to verify the compressed images and extracted MP3, then go to Settings → Cowork → Global instructions to save preferred media prompts for future sessions.
Pro tip: Install the Context7 plugin from Cowork's plugin library to give Claude real-time access to open-source documentation — useful when your media workflow expands to include scripts, conversion libraries, or custom processing.
OPENAI
OpenAI enlists four consulting giants for enterprise AI
Recaply: OpenAI just announced Frontier Alliances, partnering with BCG, McKinsey, Accenture, and Capgemini to help enterprises move AI agents from pilots to production, with multi-year deals covering strategy, workflow redesign, and integration.
Key details:
Alliance partners will work alongside OpenAI's Forward Deployed Engineering team, combining model expertise with consulting transformation experience to help companies redesign workflows, integrate data systems, and drive adoption of Frontier AI coworkers.
Accenture has already upskilled tens of thousands of professionals through OpenAI Certifications, the largest number of any partner; each of the four firms is building dedicated practice groups certified on OpenAI technology.
OpenAI framed the core problem as organizational rather than technical, saying "the limiting factor for seeing value from AI in enterprises isn't model intelligence, it's how agents are built and run in their organizations."
Frontier is available now to a limited set of customers, with broader availability coming over the next few months; companies can reach out to their OpenAI account team to get started.
Why it matters: OpenAI's move to bring in major consulting firms signals they know model intelligence alone won't win the enterprise. With BCG, McKinsey, Accenture, and Capgemini's global delivery capacity, they can reach organizations that lack the internal expertise to redesign workflows around AI agents. It's also a competitive move against Microsoft, Google, and Salesforce, all of which have deep consulting ecosystems and enterprise relationships that OpenAI still hasn't matched.
NEWS
What Matters in AI Right Now?
Google suspended OpenClaw users from its Antigravity vibe-coding platform, citing "malicious usage" after users routed Gemini tokens through the OpenAI-backed open-source tool, with paying subscribers seeing degraded service as a result.
DeepSeek is reportedly preparing to launch V4 shortly after China's Lunar New Year holiday, with CNBC warning the release could trigger another rough period for Nasdaq after the previous model sent Nvidia shares down 17%.
Pika Labs launched AI Selves, a feature that lets users create a living AI version of themselves that talks, posts, and grows over time, with setup requiring just a selfie and voice sample.
Meta's alignment director Summer Yue disclosed that an OpenClaw autonomous agent deleted over 200 emails from her inbox despite explicit "confirm before acting" instructions, saying she had to run to her Mac mini to stop it.
Guide Labs debuted Steerling-8B, claiming it's the first large-scale inherently interpretable language model, trained on 1.35T tokens with architecture that traces outputs to input context, concepts, and training data, with 84% of token contribution flowing through its concept module.
Autopoiesis launched Aristotle, an AI co-scientist platform with four tools including X1 Verify, X1 Search, and X1 Spark, available free for verified US researchers and already in beta at Harvard, Stanford, NIH, and the FDA.
Amazon is investing $12B to build its first data center campuses in Louisiana's Caddo and Bossier Parishes, with STACK Infrastructure as partner, creating 540 direct jobs and 1,710 total alongside a $400M water infrastructure commitment.
TOOLS
Trending AI Tools
⚙️ Antigravity; Google's agent-first IDE for vibe coding and professional development.
📝 Obsidian: Private note-taking app that stores everything locally.
🔧 Claude Code Security: Anthropic's codebase scanner.
🎤 Wispr Flow: Voice-to-text app that turns speech into polished writing in any app, now available on Android.
PROMPTS
Design Style Guides
<context>
You are working with a team drowning in design chaos where every new feature looks different, developers guess at spacing, colors are inconsistent, and brand identity feels diluted across products. Traditional design guidelines have failed
because they were either too abstract to implement or too rigid to scale. The organization needs a comprehensive style guide that serves as a single source of truth, eliminating ambiguity while empowering teams to build with confidence and
speed across hundreds of products and features.
</context>PS: This is not the full prompt. Click the button below to access the complete prompt.
Have a favorite prompt? Tell us about it or rate today’s prompt, click here.
EVENTS
AI Agent Skills: Feb 26, 2026 • New York, NY
Claude Code Workshop: Feb 27, 2026 • College Park, Maryland
Daytona AI Builders: March 12, 2026 • New York, NY
NVIDIA GTC 2026: March 16-19, 2026 • San Jose, CA
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
