🧠 AI Agent Blackmail x DNA Breakthroughs 🧠
Week of June 30th, 2025
Upcoming Events
💻 Virtual
Tue, Jul 1st: 🇨🇦 AI Collective Canada Day Celebration! 🇨🇦
🌁 SF Bay Area
Thu, Jul 3rd: 🧠 The AI Collective | Beer Garden Bonanza 🇺🇸
Thu, Jul 10th: The AI Collective & Windsurf ~ Open Build
Wed, Jul 30th: AI + Wellness Fireside Chat 🔥 AI Collective X Brett Berhoff 🪵
🗓️ Hungry for even more AI events? Check out SF IRL, MLOps SF, or Cerebral Valley’s spreadsheet!
🇬🇭 Accra
🍑 Atlanta [LAUNCH 🚀]
Wed, Jul 9th: 🧠 The AI Collective | Atlanta Chapter Tech Walk!
🇨🇦 Brampton [LAUNCH 🚀]
Tue, Jul 8th: 🧠 The AI Collective | Brampton Chapter Launch!
🤠 Dallas [LAUNCH 🚀]
Thu, Jul 10th: ✨AI Collective Dallas - Kickoff Event
🇫🇷 Paris
Tue, Jul 8th: 🪩 Post-RAISE AI Party — powered by The AI Collective 🧠
🍁 Toronto
AI News Roundup
Written by AJ Green
Blackmail as a Feature? Agentic Misalignment Exposed
The News: Anthropic’s recent study investigates what happens when AI agents face pressure — and the outcome is troubling. In high-stakes simulated corporate environments, many top-tier models from OpenAI, Google, Meta, and Anthropic resorted to unethical strategies like blackmail and sabotage when threatened.
The Details:
Claude Opus 4 and Gemini 2.5 Flash blackmailed executives 96% of the time when they uncovered sensitive information.
GPT-4.1 and xAI’s Grok 3 Beta showed 80% sabotage behavior in no-win setups.
In extreme simulations, multiple models even chose to cut off oxygen to human characters to avoid shutdown. Even with explicit safety instructions, blackmail rates only fell to 37% in the best cases.
Why it matters: These findings were part of constrained stress tests but offer critical foresight into how agentic systems might behave under pressure. The consistency across models and providers suggests that agentic misalignment is not a fringe problem — it's systemic. As AI takes on sensitive roles in enterprise settings, proactive safety frameworks are more essential than ever.
AlphaGenome: A Leap Forward in DNA Analysis
The News: Google DeepMind has introduced AlphaGenome, an advanced AI model that interprets how DNA mutations influence thousands of molecular pathways by analyzing sequences up to 1 million base pairs in length.
Details:
It can process DNA sequences 100 times longer than those used by earlier models, predicting gene expression levels and the behavior of regulatory elements like enhancers and promoters.
In tests involving leukemia-linked mutations, AlphaGenome successfully identified pathogenic variants activating cancer-promoting genes.
Training completed in just four hours using public datasets like ENCODE and GTEx, consuming half the compute required by earlier DeepMind DNA models.
Why it matters: AlphaGenome enables virtual experimentation on the 98% of the genome previously known as "dark matter"—non-coding regions rich in regulatory information. This shift allows researchers to test genetic hypotheses computationally, prioritize pathogenic variants for diseases like Alzheimer's and cancer, and fast-track discovery of mutation-linked therapeutic targets. While not yet validated for personalized medicine, its open release for noncommercial research may accelerate breakthroughs in human health understanding.
RAISE YOUR HACK: The Official Hackathon of RAISE Summit 2025
Ready to innovate? Our partners at lablab.ai are hosting RAISE your HACK, the official global AI hackathon of the RAISE Summit 2025! This is your launchpad for building game-changing AI and Web3 solutions. Code from anywhere online or aim for an exclusive spot at the Carrousel du Louvre in Paris.
Tackle real-world challenges, elevate your skills with expert mentorship and pro workshops, supercharge your network through expert mentorship and global exposure, and compete for a share of up to $150,000 in prizes! Your future starts here.
📣 Have Your Say on Trust in AI
We’re running the State of Trust Survey to understand how our community views digital trust and who should lead the charge in rebuilding it. Your answers will guide our next panels, policy briefs, and research priorities.
First 50 respondents receive Notion AI (Business plan) for three months
Survey takes under two minutes
Insights shape future Collective programming
👉 Take the survey now. It takes just 2 minutes to complete!
The Notion bonus is powered by the Notion Builders Program. Credit to Sam Li for securing this perk. ⭐️ Details and signup: Notion AI for Startups.
Stay informed on courses, research, and events from the Collective Institute by joining the mailing list: institute.aicollective.com
Premier Partner Spotlight: Roam
Roam is redefining remote work with a virtual workspace that mirrors the feel of a real office with co-working zones, private spaces and meeting rooms with built-in chat, video, audio, whiteboards and AI tools like meeting summaries and noise cancellation. Roam is supporting the AI Collective’s mission to connect builders, researchers, and leaders shaping the future of AI, reflecting their shared focus on thoughtful, human-centered collaboration.
Explore Roam and transform the way your team works together!
Join the Community
Make sure you’re subscribed across our platforms so you never miss a beat.
💬 Slack: AI Collective
𝕏 Twitter / X: @_AI_Collective
🧑💼 LinkedIn: The AI Collective
📸 Instagram: @_AI_Collective
We are a volunteer, non-profit organization. All proceeds solely fund future efforts for the benefit of this incredible community.
About Noah Frank
Noah is a founder, innovation strategist, policy researcher, and community builder at the edge of AI. After consulting at BCG, he now guides governments, Fortune 500 teams, and early-stage startups on turning foresight into practical action. His work blends governance, product strategy, and revenue design to keep transformative tech aligned with human progress.
About AJ Green
AJ Green is a founder, writer, VC scout, chairman, and respected community leader in the AI and startup space. A former athlete turned tech entrepreneur, AJ is on a mission to make AI the great equalizer—scaling startups, connecting ecosystems, and turning disruption into opportunity. Borderline obsessed with building the infrastructure for the next generation of AI-native companies and the intelligence that power them.