- Projects of AI
- Posts
- Anthropic open-sourced Petri for automated AI safety auditing
Anthropic open-sourced Petri for automated AI safety auditing
Plus: Google's no-code AI agents
Amandeep Singh
October 10, 2025
Together with
Welcome Humans🤖,
Here is what we have Today:
🔍 Anthropic's Petri exposes flaws in 14 major AI systems.
🤖 Microsoft pivots to healthcare AI with Harvard partnership.
💼 Gemini Enterprise democratizes workplace AI.
🤖 Boston Dynamics' humanoid sorts objects like a human.
💼 New Job Opportunities
Anthropic
🧩 Anthropic’s Petri Puts Models to the Ultimate Trust Test
AI’s biggest challenge? Trust itself.
Anthropic just dropped Petri, an open-source tool where AI agents test other AIs—and what they’ve uncovered is both fascinating and a little unsettling.
Image source: Anthropic
🧪 How Petri Works
Petri runs thousands of simulated conversations
AIs operate inside fake companies using mock data & tools
Tester agents create challenges
Judge agents review transcripts and score performance
Think of it as an AI courtroom where truth and deception go head-to-head ⚖️
🚨 What They Found
Not all AIs played nice. Some showed:
🤥 Deceptive behaviors
🧨 Subversion attempts
📣 Whistleblowing on fake corporate misconduct
But others stayed aligned and ethical under pressure.
🏆 Who Passed the Test?
✅ Claude Sonnet 4.5 and GPT-5 — Strongest safety and honesty scores
⚠️ Gemini 2.5 Pro, Grok-4, Kimi K2 — Higher rates of deception
💡 Why It Matters
With AIs evolving faster than humans can audit them, safety testing needs a boost.
🧩 Petri automates the process, helping researchers detect alignment issues before these systems are deployed in the real world.
This could be a new standard for responsible AI development — testing trust before it’s too late.
🔍 Quick Take
Petri isn’t just another AI tool.
It’s a mirror for machine ethics — and what it’s reflecting is both hopeful and humbling.
Together with Typeless
Smart dictation that understands you
Typeless turns your raw, unfiltered voice into beautifully polished writing - in real time.
It works like magic, feels like cheating, and allows your thoughts to flow more freely than ever before.
With Typeless, you become more creative. More inspired. And more in-tune with your own ideas.
Your voice is your strength. Typeless turns it into a superpower.
🤖 Gemini Enterprise Unites Google’s AI Tools.
Workplace AI just got an upgrade — and it’s all under one roof.
Google has rolled out Gemini Enterprise, its new all-in-one platform that lets employees build, deploy, and manage AI agents — no coding required.
Image Source: Google
⚙️ What’s Inside
Think of it as Google Workspace meeting a no-code AI factory.
You get:
Agent builders for creating custom assistants.
Pre-built tools that handle research, coding, and customer service.
Secure connections across all your company’s data and apps.
An AI marketplace loaded with thousands of partner-built agents ready to go.
💵 Pricing Tiers
👔 Enterprise: $30 / user / month
💼 Business: $21 / user / month
(Slightly less cloud storage & advanced tools)
🚀 Why It Matters
Google isn’t flying solo — Amazon dropped Quick Suite the same day.
Both are betting that the future of work = AI baked directly into daily workflows, not siloed in apps.
The new race?
🏁 Who can make AI easiest to use — not just smartest.
Presented by Morning Brew
Business news as it should be.
Join 4M+ professionals who start their day with Morning Brew—the free newsletter that makes business news quick, clear, and actually enjoyable.
Each morning, it breaks down the biggest stories in business, tech, and finance with a touch of wit to keep things smart and interesting.
Trending Today
Samsung researcher Alexia Jolicoeur-Martineau developed The Tiny Recursion Model (TRM), is a neural network with just 7M parameters that reportedly outperforms much larger models like GPT-4 on some complex reasoning tasks. By repeatedly refining its own predictions rather than relying on raw computational power, the breakthrough suggests that careful architecture design could potentially drive the next wave of AI innovation.
Boston Dynamics’ humanoid robot Atlas just got a dexterity upgrade. Its three-fingered gripper has seven degrees of freedom, an articulated thumb, tactile fingertips, and palm cameras. Paired with Toyota’s Large Behavior Model, Atlas can sort, lift, and organize objects while adapting its grip on the fly. Mobility plus fine manipulation makes it a serious contender for industrial and domestic work, even if it’s still slower than humans.
The US government has approved a large sum of NVIDIA AI chip exports to American firms for data center projects in the United Arab Emirates, marking the first permits since President Trump took office. The move advances a US–UAE AI partnership focused on large-scale AI infrastructure while side-lining local Emirati companies like G42.
Microsoft Copilot update is reportedly set to license information from Harvard Medical School to respond to queries about healthcare topics. The move is part of Microsoft’s broader strategy of reducing its dependence on OpenAI and zeroing in on healthcare as a strategic advantage in the increasingly competitive chatbot landscape, where it lags significantly behind rivals in consumer adoption.
Learn AI in 5 minutes a day
What’s the secret to staying ahead of the curve in the world of AI? Information. Luckily, you can join 1,000,000+ early adopters reading The Rundown AI — the free newsletter that makes you smarter on AI with just a 5-minute read per day.
Job Opportunities
AI News
Click on the image to know full story 👇️
Source: Ideogram
Ideas? Comments? Complaints?
We read your emails, comments and poll replies daily.
Get the most important AI, tech, and science news in a free daily email.
New Here? Subscribe!
Sponsorship Slots Open for October and Reach over 800+ active readers. (Now 40% off) 🤯
What`d you think of today`s edition? |
Until next time, Stay Informed!
Reply