Thursday — May 28, 2026

On Thursday, the AI industry witnessed an unprecedented show of force — not from corporate giants, but from their own employees. Over 300 current and former workers from Google, OpenAI, and other leading AI labs filed an amicus brief in support of Anthropic, which is locked in a legal battle with the U.S. Department of Defense over refusing to deploy its AI for specific military purposes. The brief argues that AI companies should have the right to refuse contracts that violate their ethical guidelines, and that employees should not be forced to build systems they believe could cause harm.

This is not an abstract debate. The Pentagon has been aggressively pursuing AI partnerships for everything from logistics to autonomous systems, and Anthropic's refusal to play ball made it a target. The lawsuit could set a precedent for whether AI companies can pick and choose their government clients based on ethical red lines. If the court rules against Anthropic, it could force every AI lab to accept military contracts or face legal consequences — a chilling prospect for builders who care about responsible AI.

What makes this story wild is the cross-company solidarity. Google and OpenAI are direct competitors with Anthropic in the AI arms race, yet their workers are publicly backing a rival. This suggests that the grassroots ethical movement in AI is stronger than corporate loyalty. For builders, this is a signal that your values matter more than your employer's stock price — and that the talent market will increasingly reward companies with clear ethical principles.

The immediate practical implication: if you're building AI products today, you need to define your own red lines before someone else defines them for you. Whether it's military use, surveillance, or content moderation, the legal and reputational risks of being reactive are enormous. Anthropic is taking the hit now, but every builder will feel the ripple effects of this case.

For those working on AI safety or alignment, this is also a reminder that technical solutions alone won't save you. The ethical stance you take publicly will determine whether top talent wants to work with you — and whether your users trust you. The amicus brief includes signatures from researchers who worked on GPT-4, Gemini, and Claude, proving that even the architects of these systems have limits.

Finally, this story is a wake-up call for founders and engineering leads. If you haven't had a team-wide conversation about what you will and won't build, do it this week. Your employees are watching, and they might just side with your competitor's team if you don't lead with integrity.

The AI talent war has officially hit the college campus. According to a Business Insider report published today, companies like OpenAI, Anthropic, and Google are offering intern salaries that rival mid-level engineering roles at non-AI companies. We're talking base salaries of $10,000 to $15,000 per month, plus signing bonuses, relocation packages, and guaranteed return offers. For a summer intern, that's a six-figure annualized comp — unheard of in any previous tech cycle.

Why the frenzy? Because AI is moving so fast that even junior talent with a few months of experience building with LLMs or fine-tuning models is considered a hot commodity. Companies are betting that today's intern is tomorrow's lead researcher, and they're willing to pay a premium to lock that relationship in early. The strategy is simple: get them in the door, show them the culture, and make it painful to leave by the time they graduate.

For builders, this changes the calculus of how you staff your AI projects. If you're a startup, you can't compete on salary alone — but you can compete on impact and ownership. Interns at big labs often get relegated to narrow tasks, while a startup can give them end-to-end ownership of a model deployment or a critical feature. That's a powerful recruiting tool.

The downside? This arms race is inflating compensation expectations across the board, making it harder for smaller teams to hire any AI talent at all. If you're bootstrapping, you might need to rethink your hiring strategy entirely — focus on remote talent, non-traditional backgrounds, or even hiring from adjacent fields like computational linguistics or data engineering.

There's also a cultural risk. Interns who are treated like royalty from day one may develop unrealistic expectations about career progression. But for now, the market is so hot that most companies are willing to take that gamble. The key takeaway for builders: start building relationships with universities and student AI clubs now. The best interns are being snapped up before they even submit their first application.

Finally, this trend underscores the importance of brand. Interns are choosing companies based on mission and ethics as much as comp — which is why Anthropic's principled stance on military AI might actually help it win the talent war, even if it can't match Google's salary bands. Builders should lean into their unique culture and values to attract the next generation of AI talent.

In what might be the most entertaining AI benchmark of 2026, Google, OpenAI, and Anthropic are pitting their flagship models against each other in a Pokémon Red/Blue speedrun competition. Each model controls a separate Twitch stream, making decisions in real-time as they navigate Kanto, catch Pokémon, battle gym leaders, and attempt to beat the Elite Four. The streams have become a spectator sport, with thousands of viewers watching their favorite AI stumble, learn, and occasionally pull off brilliant strategies.

Why Pokémon? Because it's a surprisingly rigorous test of an AI's reasoning abilities. The game requires long-term planning (which gym to tackle next, which Pokémon to train), short-term tactics (which move to use in battle), and the ability to recover from failure (when your starter faints or you get lost in a cave). It's a closed-world environment with clear rules, but the combinatorial complexity is enormous — there are hundreds of Pokémon, moves, and items to consider.

Early results show that each model has distinct strengths and weaknesses. OpenAI's model is aggressive, often overleveling its starter and brute-forcing battles, but it struggles with puzzles like the Safari Zone. Google's Gemini is methodical, exploring every route and item, but it can get stuck in analysis paralysis. Anthropic's Claude is the most cautious, often avoiding risky battles and prioritizing healing, but it sometimes misses opportunities for big gains.

For builders, this is more than a fun distraction. It's a live demonstration of how different architectures and training approaches handle real-time decision-making under uncertainty. If you're building an AI agent that needs to operate in a dynamic environment — like a customer support bot or a trading algorithm — watching these streams can give you intuition about which model behavior you prefer.

There's also a practical lesson: benchmarking matters, but creative benchmarks like this reveal failure modes that standard tests miss. A model that scores 99% on a multiple-choice reasoning test might still get stuck in Viridian Forest for hours. If you're evaluating AI for your own use case, don't rely solely on leaderboards — design a test that mimics your actual deployment conditions.

Finally, this competition highlights the growing trend of gamification in AI research. By making benchmarks public and entertaining, companies can engage the community, attract talent, and get free feedback on their models' weaknesses. Builders should consider running their own gamified tests — even a simple game like Pokémon can surface issues that would never appear in a controlled lab setting.