ME News message, April 23 (UTC+8), according to Beating monitoring, the UCSB Yu Feng team, in collaboration with fuzz.land and other organizations, proposed AgentFlow, which automatically synthesizes multi-agent harnesses (programs that orchestrate agent role assignments, information transmission, tool allocation, and retry logic) for vulnerability discovery. The paper states that when the model remains unchanged, merely modifying the harness can yield success-rate differences by several multiples, but existing solutions are mostly handcrafted or only search a limited design space. AgentFlow uses a typed graph DSL to unify the five harness dimensions (roles, topology, message patterns, tool bindings, and coordination protocol) into an editable graph program, enabling in a single step the simultaneous addition and modification of agents, topology, prompts, and tool sets. The outer loop pinpoints failure stages from runtime signals such as target-program coverage and sanitizer reports, replacing binary pass/fail feedback. On TerminalBench-2, paired with Claude Opus 4.6, it achieves 84.3% (75/89), the highest score among similar categories on this leaderboard. On the Chrome codebase (35 million lines of C/C++), the system synthesizes harnesses composed of 300+ agents; the automatically evolved agent instructions specifically target C++ memory-safety vulnerabilities, require crash verification using ASAN/UBSAN, and use multi-agent deduplication via shared documents and file locks. Running the open-source model Kimi K2.5 on 192 H100s for 7 days, it discovered 10 zero-day vulnerabilities, all confirmed by Chrome VRP. Six of them have been assigned CVE IDs, involving WebCodecs, Proxy, Network, Codecs, and Rendering; the types include UAF, integer overflow, and heap buffer overflow. Among them, CVE-2026-5280 and CVE-2026-6297 are Critical-level sandbox escapes. fuzz.land co-founder Shou Chao Fan stated that some vulnerabilities were originally discovered using MiniMax M2.5, and MiniMax M2.5 and Opus 4.6 can also find most of them. AgentFlow has been open-sourced. (Source: BlockBeats)

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

Reward
like
Comment
Repost
Share

Comment

Add a comment

No comments

Trending Topics
View More
#
Get2SharesOfSKHynixAtZeroCost
1.55M Popularity
#
MicronOvertakesMetaInMarketValue
212.66K Popularity
#
WorldCup🇫🇷vs🇳🇴
328.2K Popularity
#
USMayPCEInflationRisesTo4.1%HighestIn3Years
557.86K Popularity
#
StakeUSD1Earn9.48%APR
982.62K Popularity

Pinned

Sitemap

AgentFlow synthesizes 300 Agents to dig out 10 Chrome zero-day vulnerabilities including sandbox escape.

Trending Topics

Get2SharesOfSKHynixAtZeroCost

MicronOvertakesMetaInMarketValue

WorldCup🇫🇷vs🇳🇴

USMayPCEInflationRisesTo4.1%HighestIn3Years

StakeUSD1Earn9.48%APR

Pinned