thehackernews.com 23 Oct 2026, 08:35 UTC

Flaws in x402 SDKs let AI agents pay API bills without spending limits

SECURITY firm V12 has disclosed a set of critical defects in the x402 SDKs used by AI agents to pay for API calls. The most serious flaw stemmed from the default client, which previously allowed automatic signing of payments without any spending limit. In August, spend controls were introduced, capping payments and restricting accepted tokens to US dollar stablecoins such as USDC.

Before that fix, an adversarial API could prompt a payment for the wallet’s full balance, potentially draining funds if the agent’s app did not impose its own limits. V12 identified 25 flaws across the x402 codebase, three of which were deemed critical and described in detail; all three were addressed in the August 18 releases.

The report also details two other notable issues: a flawed facilitator in Solana payments that could accept an approval rather than a transfer as a valid payment, and a Sign-In-With-X extension flaw where a login message could be reused for access on another site. The TypeScript implementations were affected until the August updates, with the Python and Go clients already enforcing checks.

Evidence from V12’s tests showed no public attacks to date, and the X402 Foundation has not issued a dedicated security advisory for these flaws. As of 23 October 2026, latest versions were 2.28.0 for @x402/core, @x402/svm, @x402/extensions, @x402/fetch, and @x402/axios, with 2.28.0 for the Go module and 2.25.0 for Python. Upgrades to these releases are advised, and where upgrades are not yet possible, policy-based limits can still be configured.

View full article

Article by CyberSIXT