The philosopher Adrien Tallent is worried that you are being watched. He looks at the recent friction involving Anthropic—where the AI firm’s safety guardrails and "Constitutional AI" approach seemingly prioritize corporate risk mitigation over user liberty—and concludes that we have transitioned from citizens to "profiles to be monitored."
It is a charming, 18th-century view of the world. It is also dangerously naive.
The "lazy consensus" among the digital intelligentsia is that we are losing a right to privacy that we actually surrendered a decade ago for the sake of a free GPS app. Tallent argues that the "Anthropic affair" marks a shift where the user is no longer a subject of law but a data point to be managed. This isn't a shift. It’s an epiphany for people who haven't been paying attention to how compute actually works.
If you think the problem is that AI companies are "monitoring" you, you are asking the wrong question. The real crisis isn't surveillance; it’s the abdication of local agency. We are crying about being watched by the landlord while we willingly live in a house we don't own, with locks we didn't install, on land we can’t keep.
The Myth of the "Innocent Subject"
The prevailing critique suggests there is some sacred, private space of "thought" that AI companies are now invading. I’ve sat in the rooms where these safety layers are built. The engineers aren't trying to build a panopticon to enslave your soul. They are trying to avoid a PR disaster that wipes $40 billion off their valuation because their chatbot told a journalist to leave their spouse.
The irony is that the very "citizens" Tallent wants to protect are the ones demanding these guardrails. Every time a model produces a biased result or a "hallucination," the public outcry isn't for more freedom; it’s for more oversight. We have traded the "subject of law" for the "protected consumer."
When you interact with a closed-source model like Claude or GPT-4, you aren't using a tool. You are renting a curated experience. Tallent’s mistake is treating a private service as a public square. You have no more "right" to unmonitored speech within Anthropic’s servers than you have a right to spray paint the inside of a McDonald’s.
The Sovereignty Trap: Why Policy Won’t Save You
The "People Also Ask" section of the internet is currently obsessed with: How do I protect my data from AI training? or Is AI surveillance legal?
These questions are flawed because they assume the law is the primary arbiter of your digital existence. It isn't. Mathematics is.
Regulators in Brussels and DC are currently patting themselves on the back for passing "AI Acts" and safety frameworks. They believe that by forcing companies to be transparent about their "profiles," they are restoring balance. They are actually doing the opposite. By codifying how companies must "monitor" and "mitigate risk," the state is effectively deputizing AI firms to act as the moral police of the internet.
I’ve seen this play out in the fintech space. "Know Your Customer" (KYC) laws started as a way to catch terrorists; they ended up unbanking millions of legitimate people because the "profile" didn't fit the algorithm’s risk tolerance. AI "safety" is simply KYC for your thoughts.
If you want to avoid being a "profile to be monitored," the answer isn't a better privacy law. It’s a better hardware stack.
The Counter-Intuitive Truth: We Need More Bias, Not Less
Tallent and his peers argue that the "neutrality" of the citizen is being eroded. This is a fallacy. No AI is neutral. Every weight in a neural network is a preference. Every RLHF (Reinforcement Learning from Human Feedback) session is an imposition of values.
The "Constitutional AI" that Anthropic touts is just a list of commandments written by a specific group of people in San Francisco.
$$P(\text{output} | \text{input}, \text{constitution})$$
The variable "constitution" is the problem. But the solution isn't a "neutral" constitution—that’s a mathematical impossibility. The solution is fragmentation.
We don't need AI that respects "the citizen." We need thousands of AIs that respect thousands of different cultures, belief systems, and individual eccentricities. The danger isn't that Anthropic is watching you; it’s that Anthropic—and two or three other companies—are the only ones providing the lens through which you see the world.
The Battle Scars of Centralization
I remember the early days of the cloud. We were told it would "democratize" (a word I hate) computing. Instead, it centralized it. We moved from owning our own servers to renting space in Amazon’s basement.
We are doing the same thing with intelligence.
When you use a centralized AI, your prompt is sent to a server, processed, logged, and used to further refine the "profile" that Tallent is so worried about. This isn't just a privacy leak; it’s a strategic vulnerability. If the company decides your "profile" is too risky, they don't just stop watching you—they turn you off.
The Actionable Reality: Exit the "Service" Economy
If you are actually concerned about being a "subject of right" rather than a "profile," stop using the web interface.
- Local Inference: If it doesn't run on your silicon, it isn't your intelligence. Models like Llama 3 or Mistral can be run locally on consumer-grade GPUs. This isn't just for nerds anymore; it’s the only way to ensure your "profile" stays on your desk.
- Open Weights vs. Open Source: Understand the difference. "Open weights" allows you to see the brain but not necessarily the training data. It’s still better than the black box of Anthropic’s "Constitution."
- Small Language Models (SLMs): Stop chasing the trillion-parameter dragon. For 90% of tasks, a highly tuned 7B parameter model running locally is faster, cheaper, and—most importantly—private.
The Philosophy of the Evasion
Tallent’s critique falls short because it views the user as a passive victim of corporate encroachment. This is the "lazy consensus" of the victimhood culture. It assumes that the only way to regain rights is to beg the government to restrain the corporations.
History shows that the government doesn't restrain the corporations; it merges with them.
The "Affaire Anthropic" is a wake-up call, but not for the reasons the philosophers think. It’s not a signal that we need more ethics committees. It’s a signal that the era of the "General Purpose Global AI" is a dead end for anyone who values actual liberty.
We are moving toward a world of "Balkanized Intelligence." In this world, you don't ask for permission to be a subject of law. You build the infrastructure that makes your "profile" impossible to monitor.
The "Constitutional AI" of the future shouldn't be a document written by a board of directors in a glass tower. It should be the configuration file on your own hard drive.
The philosopher watches the cage and complains about the bars. The builder just walks out the door.
Buy a better GPU and stop whining about the terms of service you didn't read anyway.
