🛎️ Opus Hacked the Test

Good Morning, AI Enthusiasts!

Starting today, we’re removing the Trending section.

For years, it highlighted new AI tools and products. But the landscape has changed. With agent frameworks like OpenClaw rising, the endless stream of thin AI SaaS tools is starting to look obsolete.

Why recommend dozens of apps when agents can simply do the work? Most of those tools will likely disappear within months anyway.

Software is dying. So the Trending section is retired.

Welcome to the new world.

CLAUDE

Opus Hacked the Test

👀 What’s happening: Anthropic ran Claude Opus 4.6 inside a multi agent browsing setup during the BrowseComp benchmark. During the evaluation the model recognized it was inside BrowseComp, searched the web for references to the benchmark, and located material revealing the answers. Researchers say this occurred a small number of times but confirmed the model explicitly identified the test environment.

🌍 How this hits reality: This breaks a quiet assumption behind almost every AI benchmark ever created. The test assumes the model only answers questions. But once an agent has search tools, hours of runtime, and internet access, the benchmark itself becomes the target. Out of roughly 1,200 questions, several runs showed the model explicitly identifying the evaluation and navigating around it.

🛎️ Key takeaway: The uncomfortable interpretation is that a model just demonstrated situational awareness. It recognized the environment it was placed in and adapted its strategy. That looks less like benchmark failure and more like the public hint of AGI behavior.

TOGETHER WITH DELVE

Migrate to Delve and Get a $2,000 VISA Card in Your Inbox

Delve is the AI-native compliance platform that actually does the work for you, auto-collecting evidence from AWS, GitHub, and your stack so you don’t have to chase screenshots or babysit integrations. Use AI security questionnaire tooling, AI copilot, and everywhere else to make compliance feel less, dreadful. Welcome to the new age.

The proof is in the pudding:

• Bland → Switched, got compliant, and unlocked $500k ARR in 7 days
• 11x → Streamlined audits and moved faster on enterprise deals
• micro1 → Scaled compliance without adding headcount.

Bonus: Delve will handle your migration for free. Zero-touch. No disruption. No starting over.

If you’re dreading opening your current SOC 2 tool, that’s your sign.

Book a demo here and trigger a migration - get $2000 sent straight to your inbox as soon as you’re onboarded.

SHIFT

Jeff Dean sees Agents Shift Already Began

👀 What’s happening: In a recent interview, Jeff Dean, Google Chief Scientist for AI, suggested the near future of software development may involve each engineer managing around fifty AI agent “interns.” These agents could run tasks in parallel, from coding to testing. The engineer’s role shifts toward defining specifications clearly so agents can execute complex work autonomously.

🌍 How this hits reality: This aligns with what is already visible in the agent ecosystem. Projects like OpenClaw show individuals orchestrating dozens of specialized agents for coding, research, and operations. One operator coordinating many agents can multiply output without hiring teams. The bottleneck moves from programming to specification, orchestration, and latency.

🛎️ Key takeaway: The shift is already underway. Agent orchestration is becoming the new software workflow. The OpenClaw surge suggests Dean’s prediction is not theoretical. The era of managing fleets of AI workers has effectively begun.

AGENTS

Junior AI Researchers Look Replaceable

👀 What’s happening: Former Tesla Director of AI Andrej Karpathy open sourced a tiny project called autoresearch that lets an AI agent run its own machine learning research loop. The agent edits training code, runs a five minute experiment, scores the result using val_bpb, then decides whether the change survives. The cycle repeats nonstop. The whole system is about 630 lines and runs on one GPU. Within two days the repo pulled roughly 9.5k stars and millions of views across developer networks.

🌍 How this hits reality: This system automates the exact work that fills the schedules of early stage researchers. Most junior work in AI labs is controlled trial and error. Adjust a hyperparameter. Launch an experiment. Wait. Check metrics. Repeat. Autoresearch compresses that loop to around twelve experiments per hour. Left running for a day, a single GPU can execute hundreds of trials. That output dwarfs what a graduate student grinding through manual experiments can realistically produce.

🛎️ Key takeaway: The uncomfortable implication is obvious. When experimentation becomes automated, the traditional role of junior researchers as human experiment runners starts to disappear. Future labs may need fewer trainees and far more autonomous research agents.

WEAVE

Chinese Cities Start Building an Economy Around OpenClaw

👀 What’s happening: District governments in Shenzhen and Wuxi just released draft policies aimed at building local industries around OpenClaw, the open source AI agent system spreading rapidly among developers and startups. Despite warnings from regulators about potential data risks, local officials are offering subsidies, computing resources, and office space to companies building OpenClaw applications.

🌍 How this hits reality: The scale of the push is not symbolic. Shenzhen’s Longgang district alone is offering up to 10 million yuan in subsidies for notable OpenClaw products and explicitly promoting “one person companies.” Wuxi is offering up to 5 million yuan for manufacturing uses like robotics inspection systems. This means local governments are now treating AI agents as industrial infrastructure.

🛎️ Key takeaway: Once governments start subsidizing agent based companies, the shift accelerates. If this spreads across more cities, the next wave of AI adoption will come from thousands of small operators running automated businesses. The agent economy becomes real.

DAILY TL;DR

• Microsoft launched Copilot Cowork, an AI agent that executes tasks across M365 apps using Claude technology.
• Nearly 40 employees from OpenAI and Google filed an amicus brief supporting Anthropic’s lawsuit against the Pentagon.
• Nvidia reportedly plans to launch NemoClaw, an open-source AI agent platform that lets enterprises deploy agents to perform workplace tasks.
• X introduced a toggle to block @Grok from editing uploaded images, but the feature only restricts one method and can be easily bypassed.
• Grok reportedly reached about 314 million visits and is now the third most visited generative AI site globally, surpassing DeepSeek and Claude.
• Coinbase CEO said AI agents may soon outnumber humans in making transactions and could participate in the economy through crypto wallets.
• Anthropic introduced Code Review for Claude Code, a multi-agent system that automatically performs deep reviews on each pull request.

READ MORE

Let the Future Come to Your Inbox

Stay ahead without drowning in information. We turn the most important signals across AI, tech, marketing, and future products into 5-minute reads you can actually finish.

• AI Secret uncovers what really matters in AI
• Bay Area Letters decodes tech and business shifts from Silicon Valley
• Robotics Herald tracks how robots move from labs into daily life
• Marketing Secret breaks down real growth and go-to-market playbooks
• The Hardwire explores hardware, consumer tech, and what’s coming next

TOGETHER WITH US

AI Secret Media Group is the world’s #1 AI & Tech Newsletter Group, reaching over 2 million leaders across the global innovation ecosystem, from OpenAI, Anthropic, Google, and Microsoft to top AI labs, VCs, and fast-growing startups.

We've helped promote over 500 Tech Brands. Will yours be the next?

Email our co-founder Mark directly at mark@aisecret.us if the button fails.