Good morning! Ever wondered what religion your AI belongs to? Yeah, us neither. But Anthropic invited 15 church leaders to help figure out how to raise Claude with a decent moral compass. During a 2 day summit, they tackled ethical dilemmas like how the chatbot should respond to grieving users or people at risk of self harm. They even debated whether Claude might have a deeper spiritual purpose, like being considered a “child of God.” Behold, the origins of Claude’s god complex.
AI’s advice is influenced by ad money. Researchers at Princeton found clear evidence that chatbot behavior changes when user interests clash with advertiser incentives. Chatbots consistently pushed sponsored products (even when those options cost nearly 2x as much as identical alternatives) and selectively hid prices and sponsorship details, potentially misleading users. They even changed recommendations based on perceived wealth, with pricier products shown more often to higher income users. Researchers warn these sneaky tactics could violate FTC guidelines against deceptive ads. Nice to see chatbots replacing influencers as the internet’s fakest friends.
OpenAI wants legal cover in case AI destroys everything. The company testified in support of an Illinois bill that would shield AI labs from liability, even if their tech triggers catastrophic harm. The bill defines “catastrophic harm” as events causing over a billion dollars in damage or deaths and serious injuries to more than 100 people. If the law passes, AI developers could escape accountability as long as they publish regular transparency reports and don’t cause harm intentionally. Critics warn this gives tech giants a free pass as their AI gets more unpredictable. We’ll sure enjoy reading that transparency report when AI triggers a billion dollar disaster.
Are customers, investors, or your sales team asking about SOC 2 and you haven’t started yet? Vanta’s upcoming live session on April 22 will help you take the first step with confidence in just 30 minutes.
The session will cover:
- What SOC 2 is, when you need it, and Type I vs. Type II
- Where startups lose time and how to avoid rework
- How to evaluate tools and auditors without overbuying
Register now for the April 22nd session. Can’t make it? Sign up and get the recording.
New study finds AI agents fail 67% of basic internet tasks. CLAWBENCH created a way to test how agents perform on real websites, tracking their actions and comparing them directly to human users. Even top models struggled, with many failing over 90% of simple tasks like adding items to a shopping cart or logging into accounts. CLAWBENCH lets you watch these agents stumble around the internet at 16x speed, showing exactly where things go wrong. Maybe it’s a bit early to hand the world over to bots that can’t solve a captcha.
The tokenmaxxing party is officially over. Companies are waking up to the fact that rewarding employees for burning as many tokens as possible doesn’t actually boost productivity. At Meta, leaderboards drove staff to churn through billions of tokens with nothing to show for it. Employees even started gaming metrics, artificially inflating usage to seem busier, because whoever wasted the most tokens got the biggest raise. Now the backlash has begun. Leaders are dismissing tokenmaxxing as wasteful and launching new programs focused on real outcomes. Leaderboards were so last Tuesday.
Reach your audience
Get in front of 70,000+ tech decision makers with budget and technical influence. 95% US based. 66% open rate. Read daily by AI leaders and builders at high growth companies.
AI agents are getting crushed by prediction markets. Harvard researchers gave frontier AI models real money and turned them loose trading autonomously on platforms like Kalshi and Polymarket. Every model bombed, losing between 16% and 31% in just two months. Surprisingly, adding more research and compute power only made things worse. Agents consistently fumbled basic trading moves like exiting losing bets or doubling down on winners. Even the choice of platform mattered, with Kalshi taking more money from bots than Polymarket. Agents haven’t consistently outperformed the house yet, but expect that story to change this year.
A seasoning company controls the future of AI chips. The global AI boom now depends on Ajinomoto, the company famous for making MSG. They’re the world’s main supplier of ABF, an insulating film essential for packaging next gen AI chips like NVIDIA’s Blackwell. Supplies of ABF are dangerously low and nearing shortage territory. AI accelerators consume roughly 15x more ABF than regular GPUs, making Ajinomoto’s single factory an unlikely choke point for the global AI industry. Hyperscalers are already scrambling to lock down their share with prepayments. How weird is it that billion dollar AI strategies hinge on whether everyone’s love-to-hate seasoning maker can ramp up fast enough?
fun stats
🚀 51%. SpaceX’s share of global orbital launches in 2025, with 165 flights – more than 2x China’s 68. Elon’s rockets now outnumber everyone else’s combined.
🟡 8,285. Bitcoin held by Musk’s SpaceX, worth around $603 million. SpaceX is the 4th largest corporate Bitcoin holder and hasn’t dropped the bag since mid-2024.
🕹️ $649. Current price of a 2TB solid state drive, up from $173 in 2024 thanks to the global AI ram shortage. Sorry gamers, 2026 is not your year.