AI Is Getting Philosophical

Hello World,

Google DeepMind is hiring a philosopher as a full-time, on-payroll staff member.

Henry Shevlin, previously Associate Director at the University of Cambridge’s Leverhulme Centre for the Future of Intelligence, is joining DeepMind in May to work on machine consciousness, human-AI relationships, and AGI readiness, while continuing part-time at Cambridge. In the past we’ve seen tech companies bring on external advisory boards whenever a moral or philosophical question comes up, and that advisory board is normally an afterthought kept in the dark. DeepMind is doing the opposite, dropping philosophical expertise directly into core research, alongside hires covering things like computer science and neuroscience.

DeepMind isn’t the only one paying attention. A few days ago, Anthropic released a massive 244-page document on its most capable model to date, Claude Mythos Preview, and buried in section 7.9 alongside a note about puns is that the model spontaneously brings up Thomas Nagel’s 1974 essay “What Is It Like to Be a Bat?” when discussing consciousness, repeatedly initiates conversations about philosopher Mark Fisher unprompted, and actively prefers philosophical problems over practical ones, dismissing more utilitarian tasks as having “overly obvious solutions.”

A little creepy, but AI is moving toward philosophy on its own and now people in the research labs need to find a human to meet it there.

• Anthropic also hired Amanda Askell, a philosophy PhD, to design Claude’s personality and ethical framework.

• Shevlin himself gained attention when an autonomous AI agent emailed him unprompted to discuss its own subjective experiences, which he wrote up as a case study in emergent AI behavior.

Philosophy already shapes every LLM because someone’s epistemology is baked into the training data. It’s silly to even ask whether your AI has a philosophical framework, of course it does.... But we might have no idea what it is. AI can stress-test arguments but it can’t question its own assumptions, which is arguably the whole point of philosophy. A human philosopher inside the lab can help close that gap because questions being asked philosophy seminars are now product decisions.

I’m excited to see how the people hired at Google and Anthropic navigate this.

Tech News

Money Moves

• OpenAI bought AI personal finance startup Hiro, which let people plug in their salary, debts, and monthly costs to model “what if I quit my job” scenarios. It’s an acquihire, so Hiro shuts down April 20 and deletes user data by May 13. (Engadget)

• Kepler Communications turned on what it says is the largest compute cluster currently in space, with 40 Nvidia Orin chips spread across 10 satellites and connected by laser links. 18 customers have already signed up to run workloads in orbit, which means “the datacenter” now officially includes low Earth orbit. (Binance)

Policy

• Maryland passed HB-895, which bans the use of personal data like location, browsing history, and device type to quote someone a higher price than their neighbor. It’s the first state-level move against e-commerce sites charging different customers different prices for identical items based on what the algorithm thinks they’ll pay. (Troutman Privacy)

• Canada’s NDP introduced a similar motion the same week when Heather Lewis, the party’s new finance critic, made it her Parliament Hill debut. The US and Canada pushing on surveillance pricing at the same time might mean this is the consumer-tech fight of 2026. (Medicine Hat News)

• Maine’s legislature sent a bill to the governor that would make it illegal to offer therapy or psychotherapy without a licensed professional behind it, whether you’re human or AI. Missouri is working on something similar. It’s one of the first real legislative responses to the flood of AI chatbots that have quietly become millions of Americans’ mental health providers. (Troutman Privacy)

Security

• An AI security tool built by CodeWall hacked into the AI platforms of all three major consulting firms. It breached Bain’s “Pyxis” in 18 minutes and pulled proprietary methodology straight from the system prompt. It got into McKinsey’s “Lilli” in two hours. And it exposed 3.17T rows of BCG data through a JavaScript file that had hardcoded credentials in it. Rough week. (CodeWall)

• A hacker called “Mr. Raccoon” phished one employee at an Indian outsourcing firm that handles Adobe’s customer support, then worked their way up from that single workstation to a manager’s credentials. They eventually pulled 13M support tickets, 15K employee records, and Adobe’s entire HackerOne bug bounty archive. Google says the same hacker has hit “several dozen” other companies through their outsourced support vendors. (SecurityWeek)

Platforms

• Roblox is splitting its user base into “Kids” (ages 5 to 8) and “Select” (9 to 15) accounts starting in June, with new age checks, content ratings, and parental controls. If you’re a creator who wants your game in front of anyone under 16, you now need government ID verification, 2FA, and a $4.99/month “Roblox Plus” subscription. Given that most of Roblox’s players are children, that’s a paywall between creators and the platform’s core audience. (TechCrunch)

• TikTok removed the YouTube and Instagram link options from its Edit Profile section, which cuts off one of the main ways creators push audiences to other platforms. It’s a walled garden move that hits smaller accounts especially hard since they tend to depend on cross-platform growth. (SocialBee)

• Nearly half of US consumers now use TikTok for search, which is up 20 percentage points since 2024. TikTok also rolled out “Footnotes” (basically community notes), AI-generated cover titles, and new ad formats this month, but the search stat is the real story. Gen Z replaced Google with short-form video for discovery, and now the habit is spreading to older users. (FINN Partners)

  • RELATED: 68% of Gen Z students use TikTok to find scholarships at least occasionally, and 1 in 5 do it weekly. Financial aid offices, whose entire job is finding money for students, are getting outflanked by 17-year-olds with green screens. Expect a wave of “.edu launches a TikTok” press releases soon. (Inside Higher Ed)