
OpenAI Launches Daybreak Cybersecurity Initiative; Crypto Analysts Draw Lessons on Pre-Exploit Defense
OpenAI launched Daybreak on May 11, a cybersecurity initiative pairing its AI models with the Codex coding agent to find and validate software vulnerabilities before exploitation. CryptoSlate argues the approach offers a template for crypto, where TRM Labs data shows $2.87 billion stolen across roughly 150 hacks in 2025.
Written by CoinArticle’s AI Newsroom · from 2 cited sources. How we work
What OpenAI Announced
OpenAI launched Daybreak on Monday, May 11, a cybersecurity initiative that combines its AI models with Codex, its coding-focused agentic system. According to the company, the tool helps security teams review code, analyze dependencies, model threats, validate patches and investigate unfamiliar systems, with the stated goal of shortening the time between identifying a vulnerability and fixing it.
CEO Sam Altman described the effort in a post on X as an "effort to accelerate cyber defense and continuously secure software." OpenAI did not respond to a request for comment from Decrypt. The company said it plans to work with government and industry partners before deploying more cyber-capable models.
Why AI Firms Are Moving Into Security
The launch follows a broader push by AI companies into security and software engineering. Anthropic has marketed its Claude models for coding and security tasks, and Decrypt reports that Mozilla said it used Claude Mythos, released last month, to find 271 unknown vulnerabilities in Firefox.
The same capabilities raise concern about misuse. Researchers and government agencies have warned that advanced models could speed up vulnerability research, malware development and exploit creation. Google researchers have said large language models are getting better at finding weaknesses that conventional scanners miss. Experts remain divided on how large the threat is.
Applying the Model to Crypto
CryptoSlate argues that Daybreak's emphasis on continuous, pre-exploit security is relevant to crypto, where the usual pattern is reactive: audit before launch, monitor after deployment, respond once funds move. The outlet's central point is that by the time a bug surfaces in that model, capital has already left the protocol.
The article cites TRM Labs' 2026 Crypto Crime Report, which put 2025 illicit thefts at $2.87 billion across nearly 150 hacks and exploits. Of that, $2.2 billion came from infrastructure attacks involving compromised keys, wallet systems, privileged access, front ends and control planes. Code exploits, the category audits most directly address, accounted for $350 million, or 12.1%.
Limits of Audit-Centric Security
CryptoSlate cites Hacken data showing Web3 lost $482 million across 44 incidents in the first quarter. Six of those incidents involved audited protocols, one of which had undergone 18 separate audits. A $282 million theft involved no code exploit at all; the attacker bypassed the contract layer and compromised surrounding operational and social infrastructure.
CertiK's wrench-attack tracking adds a physical dimension. It recorded 34 verified physical coercion incidents from January through April 2026, up 41% from the same period in 2025, with estimated losses of about $101 million. Extrapolating that pace, CertiK estimates around 130 incidents for the full year. CryptoSlate notes that the targets are now often the people holding keys, including multisig signers and engineers with cloud console access.
Data Sources Differ on Scope
The figures come from different methodologies and do not fully reconcile. TRM counts $2.87 billion for 2025 across a full calendar year, while Hacken's $482 million covers only the first quarter of 2026, so the two are not directly comparable. CertiK's wrench-attack estimate covers a four-month window and is a projection, not a closed total. Readers should treat the 2026 extrapolation as an estimate from one firm.
CryptoSlate's recommendations, including continuous AI-assisted code review, threat modeling across upgrades, oracle and dependency analysis, patch validation before governance execution, and regular privileged-access reviews, are the outlet's own proposals. Neither OpenAI nor the crypto security firms cited have said Daybreak is intended for or tested on blockchain systems.
Why It Matters
For Traders
Crypto security data shows most 2025 losses came from key and infrastructure compromises rather than code bugs, so exposure tied to multisig and access controls deserves attention.
For Investors
Growing demand for AI-driven security tooling could create a new category of vendors and partnerships across both traditional software and blockchain infrastructure over the coming year.
For Builders
Protocol teams should review privileged-access, signer and front-end controls as closely as contract code, since most recent losses originated outside the smart contract layer.
This article is for information only and is not financial advice. Read the full disclaimer.






