


AI + human hybrid security systems for the cyborg era. Trusted by @Uniswap, @YearnFi, @yAuditDAO, @Kinetiq_xyz and more!


A pool that becomes unusable after 300 days with no maintenance call may be technically possible. It may also be irrelevant if the maintenance function is permissionless, economically incentivized and normally called every hour. How we check AI findings:
Zero Cool @ZeroCool_AI Sep 8 Article How we reduce false positives in AI security reviews If you run security reviews with AI tools, you can definitely tell how often they turn unverified edge cases, intentional design decisions, or incomplete attack paths into noisy findings that


Zero-knowledge circuits don't operate on integers. A 256-bit value split into sixteen 16-bit limbs only holds if the circuit forces each limb into range. The fix routes MEMCOPY through the same range check, and it's implemented.
Zero Cool @ZeroCool_AI Apr 28 Article BigInt MEMCOPY allows out-of-range limbs due to missing range checks This finding was identified by Zero Cool and confirmed as a High-severity vulnerability on the ZKsync OS bug bounty program on Immunefi. Here's what we found: in the bigint_with_control circuit


Zero Cool found a High in ZKsync's Airbender circuits, paid out as a $20K bounty through @immunefi. MEMCOPY was left out of the shared 16-bit range check. With a 0xFFFF limb plus a carry of 1, the circuit accepted 65536 as a valid limb.

Recent results from @sherlockdefi Audit Engine: Ripple XRPL Lending: 1st, solo High, fastest full review (9.3 hours) CRX FX: 1st, 9 of 11 accepted Velodrome Aero: 3rd among AI tools, 11 of 17 accepted, 5 neutral, 1 dismissed Full XRPL write-up:
Zero Cool @ZeroCool_AI Sep 30 Article Zero Cool Takes First Place in Sherlock's Ripple XRPL Lending Protocol Engagement When Sherlock announced the AI-only Audit Engine review for Lending Protocol V1.1 for the XRP Ledger, we were eager to see how Zero Cool would hold up against the codebase. The engagement was not



Green PoC. Before you call it a finding: Apply the fix and rerun. Still green means it never touched the bug. Read the test body, not its name. Did a mock hand the attacker state mainnet would refuse? Fork it and run on live data.

A reportable finding names: - the code-level flaw - the actor who can trigger it - the non-attacker victim or violated protocol property - the unexpected state transition - the concrete impact
Zero Cool @ZeroCool_AI Sep 8 Article How we reduce false positives in AI security reviews If you run security reviews with AI tools, you can definitely tell how often they turn unverified edge cases, intentional design decisions, or incomplete attack paths into noisy findings that


If you're interested in using Zero Cool to secure your protocol, reach out now to request a demo (DM is open!) or visit us at zerocool.ai

Tests that cover entry but not exit, creation but not closure, or queueing but not execution have the same blind spot. In its first six months, our Test Terminator skill contributed to 84 verified findings: 10 Critical, 36 High, 38 Medium.
Zero Cool @ZeroCool_AI Aug 11 Article Test Terminator: 84 verified findings from reviewing tests Test Terminator is a new Zero Skill from Zero Cool. Skills are one of the ways we hunt for vulnerabilities, and every one of them runs on the same engine we use in our own reviews. Test Terminator


2/ Another suite seemed to cover a full lifecycle, but its setup kept the system above a minimum boundary. No test ever let the final position exit. The final transition behaved differently in production.

1/ One suite excluded a small-value state from its test environment. The suite passed, but production still allowed that state. The omitted path held an accounting failure. The test setup had erased the exact state where production broke.

Two bugs we found by reading what a test suite left out. 🧵


Congratulations to @Enter_Elysium and the @kinetiq_xyz team on the mainnet launch. We're proud to work together to help secure the ecosystem.
Elysium mainnet is now live. elysium.kinetiq.xyz

Teams building on Elysium may apply for @ZeroCool_AI AI audit credits, provided by @KinetiqFND Each audit includes a fix review. Security above all. e/acc
Zero Cool @ZeroCool_AI Nov 6 Article Hyperliquid's new L2, Elysium, now secured by Zero Cool Kinetiq recently announced Elysium, its Layer 2 solution for Hyperliquid. Teams building on Elysium can get Zero Cool AI audits at little or no cost. Why Elysium Hyperliquid keeps HyperEVM blocks
