Consider the moment when a single corporation announces it will embed artificial intelligence into the daily workflows of 50,000 employees. That is exactly what Teleperformance, the world’s largest business process outsourcing (BPO) firm, did last week. The news was framed as a leap in operational efficiency—a tool to reduce costs and improve response times. But for anyone who has spent years observing the subtle erosion of agency in centralized systems, this is not progress. It is a warning. Teleperformance’s AI rollout is a microcosm of a broader, troubling trend: the concentration of decision-making power into opaque, profit-driven algorithms that serve the corporation, not the worker or the customer.
Context
Teleperformance is the backbone of global customer service. Its 50,000 employees handle calls, chats, and emails for banks, insurance firms, tech giants, and healthcare providers. The BPO industry thrives on low-cost labor arbitrage, mostly in countries like the Philippines and India. Now, AI is poised to replace the most repetitive tasks: answering FAQs, resetting passwords, processing returns. But the devil is in the design. Teleperformance’s approach is proprietary and centralized—the AI model is controlled by the company, trained on sensitive customer data, and deployed without transparency. There is no community oversight, no user-controlled audit trail, no way for a worker to know why an AI flagged their performance or for a customer to verify that their data was not misused. This is the opposite of decentralization.
Core Insight
From a technical and values perspective, Teleperformance’s move exemplifies everything wrong with AI adoption in the 2020s: it prioritizes efficiency over dignity, speed over fairness, and profit over privacy. Let me break this down using the lens I apply to any cryptographic system—game theory and incentive alignment.
Why Centralized AI Fails the Worker
The AI system at Teleperformance will likely include performance monitoring: measuring how many calls an agent handles per minute, sentiment analysis of their voice, and even predictive models to flag potential turnover. This is not hypothetical; many BPO firms already use such surveillance. But when the AI is a black box, the worker has no recourse. They cannot challenge a low score because they do not know the algorithm’s weights. In a decentralized system, a worker’s identity and credentials could be stored on a blockchain, with zero-knowledge proofs verifying performance metrics without revealing personal data. The worker would own their reputation, not the corporation. But Teleperformance’s model ensures the opposite: total vendor lock-in.
Why Centralized AI Fails the Customer
Customer data—conversations, purchase history, financial details—will flow into Teleperformance’s AI training pipeline. Under a centralized architecture, there is a single point of failure for data breaches, a single entity with access to everything. In my experience analyzing DeFi exploits, the root cause is almost always a central authority with too much power. Here, the risks are identical. A malicious insider or a compromised API could leak millions of records. Decentralized identity solutions, like those built on Ethereum or Polkadot, allow customers to selectively disclose data via cryptographic proofs. They could prove they are over 18 without revealing their birth date, or prove they paid a bill without exposing their bank account. Teleperformance’s centralized design makes such granular control impossible.
Why the Values Clash with Decentralization
I have audited DAO grant programs and seen how centralized committees inevitably become captured by personal relationships. Teleperformance’s AI governance is no different—it is a small group of executives deciding what is fair, what is private, and what is efficient. There are no checks and balances, no mechanism for workers to vote on how AI is applied, no way for customers to opt out of model training. This is the antithesis of the Web3 ethos. The real innovation here would be to use a public blockchain to record the AI’s inference decisions and performance metrics, allowing independent auditors to verify fairness. But Teleperformance has no incentive to do so; transparency would constrain its ability to extract value.
Technical Signals I Observe
Based on my work in Web3 communities and my mathematical background, I see three immediate red flags:
- No transparency on AI model lineage. If Teleperformance uses a fine-tuned GPT-4o or Llama 3, they will not disclose it. Without open-source weights or verifiable proofs, customers cannot trust that the AI is not biased against certain accents or demographics.
- No user-controlled data sovereignty. The typical BPO contract grants Teleperformance broad rights to use interaction data for “improving services.” This is a license to train a proprietary model on sensitive data without customer consent. In a decentralized world, each interaction would be signed by the customer’s wallet, and consent could be revoked at any time.
- Lock-in to cloud providers. Teleperformance will likely use Azure OpenAI or similar. That means Microsoft could theoretically access all data, or raise prices arbitrarily. A decentralized AI marketplaces, such as those built on top of Bittensor or Akash, would distribute compute across independent nodes, preventing any single provider from holding a monopoly.
Contrarian Angle
Many in the AI community will argue that centralization is necessary for efficiency—that decentralized AI is too slow, too expensive, or too complex. They point to the latency requirements of real-time customer service and claim that blockchain-based identity verification adds overhead. I agree that today’s decentralized infrastructure is not ready for 50,000 concurrent AI agents processing millions of queries. But that does not mean the current path is sustainable. The contrarian truth is that Teleperformance’s approach will create long-term fragility. When a single update to the AI model causes a 10% drop in customer satisfaction—as has happened with many centralized LLM deployments—the company has no fallback. There is no parallel system, no community governance to roll back a bad change. In contrast, a DAO-governed AI protocol could use quadratic voting to approve new model versions, and anyone could fork the model if they disagree. Efficiency without resilience is a house of cards.
Skepticism of “Blockchain AI” Hype
Let me be clear: I am not suggesting every BPO firm should immediately deploy a blockchain AI. Many projects claiming “decentralized AI” are just Ethereum wrappers with no real decentralization. But the problem is not technology; it is values. Teleperformance could start small: pilot a decentralized identity layer for workers to own their performance records, or use a public ledger to log when AI makes decisions affecting people. That would not require 50,000 simultaneous transactions; it could be batched or use zero-knowledge rollups. The barrier is not technical infeasibility, but corporate reluctance to surrender control.
Takeaway
The Teleperformance announcement is not just a business story; it is a philosophical fork. One path leads to a world where a handful of corporations own the AI that governs labor and customer relations—a world of surveillance, data exploitation, and power asymmetry. The other path, built on decentralized identity, on-chain governance, and open AI protocols, restores agency to the individual. Which path will we choose? As I often tell my community, trust is the only native currency. If we trust a centralized BPO giant with our data and our livelihoods, we are spending that currency unwisely. The bull market euphoria around AI stocks may blind us to these risks, but the code of our future society is being written now. Let us write it with decentralization at the core.