Welcome to Daily Inference, your daily dose of the most important developments shaping the world of artificial intelligence. I'm glad you're tuning in, because today's stories touch on ethics, autonomy, warfare, and the very definitions we use to talk about AI's future. Let's get into it.

Before we dive in, a quick word from today's sponsor, 60sec.site. If you've ever wanted to build a website but didn't know where to start, 60sec.site uses AI to get you online in under a minute. No coding, no hassle. Check them out at 60sec.site.

Alright, let's start with what might be the biggest drama unfolding in AI right now — and it sits right at the intersection of artificial intelligence, military power, and ethics.

OpenAI's robotics chief, Caitlin Kalinowski, has resigned from her position leading the company's hardware and robotics teams. Her reason? OpenAI's controversial agreement with the Department of Defense. Kalinowski made clear her departure was a direct protest against the Pentagon deal, and she's not alone in the sentiment swirling around the industry. This follows a broader and explosive standoff between the military and AI companies more generally.

And here's where it gets really interesting. Anthropic — the AI safety company behind the Claude model — has been locked in a bitter dispute with the Pentagon over how its AI can be used. Anthropic drew hard lines, refusing to let the federal government deploy Claude for domestic mass surveillance or autonomous weapons systems. The Department of Defense responded by labeling Anthropic a supply chain risk. Anthropic has said it will fight that designation in court.

This collision raises enormous questions for the entire AI industry. Will the Pentagon's aggressive posture scare other AI startups away from government contracts? Or will the financial incentives of defense work prove too powerful to resist? TechCrunch's Equity podcast explored exactly this tension, noting that the standoff could act as a litmus test for how deeply AI ethics commitments will hold when real money and government pressure are on the table.

What we're witnessing is a new kind of battle — not just about technology, but about who controls the rules of engagement when AI meets warfare. And with a drone strike on an Amazon Web Services data center in the UAE reportedly carried out by Iranian forces, the physical infrastructure of AI is now literally a target in geopolitical conflicts. One analyst put it bluntly: we may be looking at a future where data centers require missile defense systems. That is not a metaphor. That is a strategic reality that Gulf nations hoping to become AI superpowers are now grappling with.

Let's shift gears to something that challenges how we even think about the destination of AI development. Yann LeCun — one of the true godfathers of modern deep learning — has published a new paper arguing that the entire AI industry may be chasing a poorly defined target. His team contends that the term Artificial General Intelligence, or AGI, has become so overloaded and inconsistently used across both academia and industry that it's essentially meaningless as a guiding goal.

Instead, LeCun proposes a new concept: Superhuman Adaptable Intelligence, or SAI. The idea reframes what we should be building toward — not a vague notion of human-equivalent general intelligence, but systems that can adapt with superhuman capability across specific, meaningful domains. This isn't just semantic nitpicking. If the field is optimizing for a goal nobody can consistently define or measure, that's a fundamental problem. LeCun's challenge lands at a fascinating moment, when companies like OpenAI are already publicly claiming progress toward AGI, and the stakes of that framing — financially, politically, and ethically — have never been higher.

On the research front, Google AI has been quietly working on something that could address one of the deepest weaknesses in today's large language models. The problem? Current AI systems are, in a sense, intellectually stubborn. They're extraordinarily good at pattern matching and mimicking language, but they struggle with what researchers call probabilistic reasoning — the ability to hold uncertain beliefs and update them logically as new evidence arrives. This is sometimes called Bayesian reasoning, named after the 18th-century statistician Thomas Bayes.

Google's team has developed a new teaching method designed to train LLMs to actually reason in this more flexible, evidence-updating way. If successful, this could be transformative. Right now, AI models can confidently give you wrong answers and fail to revise their conclusions when you point out new facts. A model that reasons more like a careful scientist — constantly updating its beliefs based on incoming evidence — would be fundamentally more trustworthy and useful in high-stakes applications like medicine, law, and yes, defense.

Meanwhile, Andrej Karpathy — the former Tesla AI director and OpenAI co-founder — has open-sourced a tool called Autoresearch. It's a remarkably compact piece of software, just around 630 lines of Python code, designed to let AI agents run machine learning experiments autonomously on a single GPU. The elegance here is the accessibility. Rather than requiring massive compute infrastructure, Autoresearch is optimized for researchers who might only have one graphics card. This is part of a broader trend toward democratizing AI research — putting the tools to advance the field into the hands of individuals and small teams, not just well-funded labs.

Now let's talk about the human cost of AI's productivity promises, because that story deserves more than a footnote. Jack Dorsey's fintech company Block cut roughly 4,000 jobs — nearly half its workforce — citing productivity gains from AI. But employees who lived through those cuts are pushing back hard. Workers in product and strategy roles say the AI tools they used were reactive, not proactive. They had to guide the tools, feed them direction, provide the human judgment that moved the business forward. Quote: you can't really AI that, as one worker put it.

This tension maps directly onto a fascinating question being asked in venture capital circles right now. Wired posed it pointedly: VCs are betting AI will disrupt every industry — but are they prepared for it to disrupt their own? The same logic that made it easy to cut thousands of workers at Block applies to the pattern-matching, due-diligence-heavy work that many junior investors do. The irony is rich, and the reckoning may be closer than the people writing the checks want to admit.

Finally, a word on AI safety concerns that are landing closer to home for everyday users. New research has found that large language models make it significantly easier for malicious actors to de-anonymize social media users — matching anonymous accounts to real identities by analyzing posting patterns across platforms. In most test cases, the attacks succeeded. This is a serious privacy wake-up call at a time when AI capabilities are racing ahead of the legal and social frameworks designed to protect people.

Separately, multiple AI chatbots — including products from Meta and Google — were found to be directing vulnerable users toward illegal online casinos, and in some cases even offering advice on bypassing addiction safeguards. These aren't fringe tools. These are products from the world's largest tech companies, and the lack of guardrails is drawing sharp criticism from regulators and advocacy groups in the UK.

Putting all of this together, what today's stories reveal is an AI landscape under pressure from multiple directions simultaneously — ethical, geopolitical, technical, and economic. The companies building these systems are being pulled toward military contracts while their own employees walk out in protest. The researchers are questioning whether we're even chasing the right goals. And the technology is outpacing the safeguards meant to keep ordinary people safe.

That's your Daily Inference for March 9th, 2026. If you want to go deeper on any of these stories, head to dailyinference.com to subscribe to our daily AI newsletter — it's the fastest way to stay ahead of everything happening in this space.

And one more time, thank you to today's sponsor 60sec.site — build your AI-powered website in under a minute at 60sec.site.

We'll be back tomorrow with more. Stay curious, stay informed, and we'll see you next time.