Helpful AI brokers require personal entry, exterior inputs, and the power to behave.
Dai argues that AI is now not restricted by functionality, however by belief.
Researcher Wei Dai—totally different from the eponymous cryptographer writer of the b-money protocol cited by Satoshi Nakamoto within the Bitcoin whitepaper—warned this June 11 that agentic synthetic intelligence is creating a brand new class of belief issues in digital methods.
Based on Dai, the origin of the issue just isn’t technical however structural: the identical properties that make an AI agent helpful are those who make it inclined to exploitation.
The skilled printed his warning within the context of a clip distributed by 1kx, a enterprise capital fund specialised in decentralized networks wherein he serves as a Analysis Companion.
Within the video, the researcher explains that An AI agent wants three circumstances to function successfully: entry to non-public info, entry to untrusted exterior enter, and the power to behave autonomously.
He provides that when these three properties coexist on the identical system, there may be all the time the chance that the agent will likely be injected utilizing malicious prompts from exterior inputs. Then I might exfiltrate confidential info or act in a dangerous method inside company methods.
The researcher, who’s the writer of greater than a dozen educational papers on safety and cryptography, offers these three properties the title “deadly trifecta for AI brokers”. He attributes the time period to researcher Simon Willison.
Dai’s warning comes days after Anthropic launched Claude Fable 5, the primary mannequin of the Mythos household out there for common use, with mechanisms that block cybersecurity queries as excessive danger.
The transfer illustrates the stress Dai describes: the offensive functionality of superior fashions grows on the similar charge as their usefulness.
Belief is a bottleneck
Dai argues that agentic AI is now not restricted by its technical capability, however somewhat for belief and safety. For the researcher, reaching the complete potential of autonomous brokers requires new belief infrastructures all through the technological stack. Though it doesn’t specify in its publication what these infrastructures can be or which actors ought to develop them.
1kx developed in its thesis “Value of Belief 2.0”, printed in June 2026, the argument that AI is appearing as an accelerator of digital belief issues.
Based on the agency, Generative AI collapsed the price of producing faux credentials, voices, counterparts and identitieswhich generated a disaster of authenticity and verifiability.
For its half, agentic AI, the doc provides, exposes new surfaces the place belief is essential and fragile: autonomous brokers require full entry to paperwork, accounts and communication channels to function successfully.
This can be a imaginative and prescient that generates controversy within the sector. Whereas some consultants assist the concept autonomous AI brokers symbolize an inevitable structural danger, others argue that the issue just isn’t the expertise itself, however the shortage of controls and human verification.
Decentralized networks as a solution to the issue
1kx, which was based in 2018 with the thesis that decentralized networks can scale back belief prices in markets the place conventional intermediaries extract rents for being reliable, believes that These networks are the one infrastructure able to fixing belief issues that agentic AI generates.
The corporate, which in 8 years gathered greater than USD 400 million in funding exits by greater than 160 firms and protocols, highlights 4 properties that centralized methods can’t replicate concurrently and who assist that argument:
- Programmable peer-to-peer settlement
- Publicly verifiable standing
- Structural neutrality
- Participation with out permissions.
The fund maintains that any centralized platform can undertake a type of properties in isolation. Nevertheless it can’t undertake all 4 concurrently with out turning into, in impact, a decentralized community.
For 1kx, that mixture of properties is exactly what the agentic AI stack requires to function with belief on a world scale, and represents the subsequent era of trusted markets that the decentralized ecosystem is positioned to seize.
Though there are critics who query this concept of decentralization as a common answer, it might underestimate actual challenges comparable to scalability, fuel prices in autonomous brokers, and already identified vulnerabilities in good contracts.
The instant problem is to advance interoperability requirements between AI-based brokers and decentralized protocols, earlier than mass adoption turns these theoretical dangers into actual losses.
