Home
Recommended
Other Links
Hacking Lab Hacking Lab Home People Publications Contact Light Dark Automatic QueryHouse: Cross-DBMS Differential Testing with LLM and Query Transpilation Seoyeon Oh , Juhyun Song , Jaemin Park , Kwanghee Lee , Minae Hyeon , Minji Kim , Sehyuk Ka , Gwangwun Jung , Brian Pak , Insu Yun September...
Automatic Program Repair (APR) has emerged as a critical technology for autonomously addressing software vulnerabilities. While recent advances in Large Language Models (LLMs) have enabled sophisticated agentic APR systems, existing approaches still struggle to generate patches for vulnerabilities whose root causes are difficult to identify. This happens because prior methods (1) merely determine patch locations based on given crash reports or (2) employ poorly engineered context, which distracts the model and hinders effective reasoning. To address these challenges, we present Prism, a multi-team LLM-based APR system. Prism employs a multi-team architecture with three specialized teams coordinated through hierarchical context management: the Analysis Team systematically explores codebases and synthesizes repair strategies, the Patch Team translates strategies into concrete patches with pre-execution validation, and the Evaluation Team executes patches and generates feedback. To systematically explore codebases, Prism supports progressive code retrieval that combines top-down structural exploration with bottom-up, query-driven search. We evaluated Prism on 92 real-world vulnerabilities from DARPA’s AI Cyber Challenge (AIxCC). Prism successfully fixed 77 bugs (83.7%), outperforming baselines by 27 to 70 percentage points. Among those, we further analyzed 21 hard-to-localize vulnerabilities where fix locations do not appear in stack traces. In these cases, Prism fixed 13 of 21 bugs (62%), which is 24 to 62 percentage points higher than other baselines. These results demonstrate the effectiveness of Prism’s design, which leverages a multi-team architecture with strategy-based patch generation and progressive code retrieval.
"wunderwuzzi's blog"
"Hugging Face disclosed an intrusion that, according to them, was driven end to end by an autonomous AI agent system.\nAlong similar lines, Sysdig recently …"
A cross-team effort to make browser image decoding safer without slowing the web down.
TALOS-2026-2426
CVE-2026-58613
A use-after-free vulnerability exists in the CldiStreamCompleteRequest functionality of Windows Cloud Files Mini Filter Driver (version(s): 10.0.26100.8457 (WinBuild.160101.0800)). A specially crafted sequence of Cloud Filter API calls can lead to privilege...
"This is a follow-up to my previous Terminal DiLLMa research, and there is a positive outcome: Apple fixed a macOS Terminal behavior that enabled a DNS-based …"
TALOS-2026-2443
CVE-2026-50475
An out-of-bounds pointer offset vulnerability exists in the NsipGetAllInformationProviderParameters function of the Microsoft Windows NETIO.sys driver. A specially crafted I/O request packet (IRP) can cause disclosure of sensitive information.
The versions below...
For years, the cyber security industry tracked AI as a force multiplier: something that made existing attack techniques faster, cheaper, and more accessible. That framing was accurate. But the Annual AI Security Report 2026 from Check Point Research documents a transition that goes further. AI has crossed from assistant to operator. Where it once helped attackers prepare, it now runs the […]
### Summary
The vulnerability bug chain consists of a verification bypass, an internal redirect, and an open redirect affecting PyPI (web service) and WebOb. This series of vulnerabilities undermi...
In this excerpt of a TrendAI Research Services vulnerability report, Yazhi Wang and Jonathan Lein of the TrendAI Research team detail a recently patched remote code execution bug in the Windows HTTP protocol stack. Successful exploitation of this vulnerability can result in a denial-of-service condi
The SQL Server Unicode problem: why your data might not be what you
Cisco Talos’ Vulnerability Discovery & Research team recently disclosed three vulnerabilities in WolfSSF, fourteen in GeoVision, and one vulnerability in VTK-DICOM.
The vulnerabilities mentioned in this blog post have been patched by their respective vendors, in adherence to Cisco’s third-party vulnerability disclosure policy.
For Snort coverage that can detect the exploitation of these vulnerabilities, download the latest rule sets from Snort.org, and our latest Vulnerability Advisories are
I’m starting a new series of blogs called “Yarden documents random Windows things” (I am open to alternative name suggestions) where I’ll write about some features and changes in Windows that knowledge of exists only in the communal brain of security researchers, in a footnote in a blog...
TALOS-2026-2420
CVE-2026-57256
A code execution vulnerability exists in the Javascript checkbox CBF\_Widget functionality of Foxit Reader (version(s): 2026.1.1.36485). A specially crafted malformed file can lead to remote code execution. An attacker can provide a malicious file to trigger this...
Key Points Introduction Since early 2026, Check Point Research (CPR) has tracked a new modular command-and-control framework used by Cavern Manticore, an Iran-nexus APT group primarily targeting Israeli organizations, with a focus on IT providers, and government sectors. Cavern Manticore is an Iran MOIS (Ministry of Intelligence and Security)-linked actor, with links to the OilRig […]
KYC: Bypass age verification using generative video models
We’re back, melting - we’ve tried shouting, screaming, and throwing things at the Sun, and it is just not working.
Before we begin our analysis, we want to be clear - given the number of vulnerabilities fixed (and some not mentioned..), we’ve struggled to have confidence in our attribution of “vulnerability <> specific CVE ID”.
We’ve performed some informed, uninformed, random guesses - but as usual, please resist the urge to send us emails explaining how awful/wrong we are. We know some of yo
TALOS-2026-2410
CVE-2026-5263
A improper input validation vulnerability exists in the X.509 registeredID name constraints enforcement functionality of wolfSSL (version(s): 5.9.1). A specially crafted X.509 certificate chain can lead to bypass registeredID name constraints. An attacker can present...
Exploring cross-domain & cross-forest RBCD: part 2
These six free settings will not make your project unhackable. Nothing will. What they will do is close the easy doors. Turn these on, and your project will be meaningfully harder to attack than it was before.
Research by: Alexey Bukhteyev Key Takeaways Introduction Over the past several years, large language models have reshaped software development, and malware development has followed the same path. Check Point Research has documented this trend from early experiments showing that AI systems could generate offensive components, to cases of cybercriminals using ChatGPT to create malicious tools, and […]
Recent advances in Large Language Models (LLMs) have enabled agentic systems for complex, multi-step tasks; cybersecurity is emerging as a prominent application. To evaluate such agents, researchers widely adopt Capture The Flag (CTF) benchmarks. However, current CTF benchmarks reuse existing challenges, which exposes them to data contamination and potential cheating. Notably, we confirmed these issues in practice by integrating web search tools into an existing agent. To address these limitations, we present CTFusion, a streaming evaluation framework built on Live CTFs. To achieve this, CTFusion preserves per-agent independence under a single team account and reduces competition impact by forwarding only the first correct flag per challenge. Moreover, we implement CTFusion as a Model Context Protocol (MCP) server on the widely used CTFd platform, which offers broad applicability to diverse CTF events and agent types. Through experiments with three LLMs, two agents, and five Live CTFs, we demonstrate that existing CTF benchmarks can be unreliable in assessing LLM-based agents, while CTFusion can serve as a robust solution for evaluating cybersecurity agents. We release CTFusion as open source to foster future research in this area.
TALOS-2026-2408
CVE-2026-6678
A integer underflow vulnerability exists in the PKCS#7 OtherRecipientInfo functionality of wolfSSL (version(s): 5.9.1). A specially crafted malformed ASN.1 record can lead to a heap buffer overflow. An attacker can arbitrary code execution to trigger this...
TALOS-2026-2409
CVE-2026-7532
A improper input validation vulnerability exists in the X.509 iPAddress name constraints enforcement functionality of wolfSSL (version(s): 5.9.1). A specially crafted X.509 certificate chain can lead to bypass of iPAddress name constraints. An attacker can present a...
Caught in the Octopus Trap: Unauthenticated RCE in Argo CD with CodeQL
Well, well, well - once again, the cat has dragged us in and spat us out.
Today, we find ourselves questioning the reality we sit within. Must it be so predictable, and why us? “But watchTowr, what do you mean?”
Well, if you’re here, you likely fit into one of the following categories:
* A dear reader,
* A group therapy accomplice
* A Groundhog Day fan club member
Why? Because we once again find ourselves talking about Citrix NetScalers. Yes, that’s right, we’ve found another excuse to cr
Security audit of the Symfony YAML component, a PHP library to parse and dump YAML. Facilitated by the Open Source Technology Improvement Fund (OSTIF) and performed by Shielder.
Completing Compliance with Evidence : A Bottom-Up Approach to NIS2,
Welcome back to another watchTowr Labs blog post.
This time, we're looking at Progress Kemp LoadMaster, a load balancer that sits at the edge of a lot of enterprise networks. Edge appliances have a habit of becoming the way in rather than the thing keeping people out, and CVE-2026-8037 keeps that streak alive: a pre-authentication Remote Code Execution vulnerability accessible to anyone who can access the API.
So, in probably a predictable turn of events, we're back doing what we do best.
As
The GitHub Advisory Database is processing more vulnerability reports than ever before. Here's what's driving the surge, how we're responding, and how the community can help.
Charting your way in: Helm template injection
TALOS-2026-2366
CVE-2026-22879
A heap-based buffer overflow vulnerability exists in the vtkDICOMItem::FindDataElementOrInsert functionality of vtk-dicom (version(s): 9.5.2). A specially crafted DICOM file can lead to heap-based memory corruption. An attacker can provide a malicious file to...
The attacker flash-borrowed mOCEAN, cycled it through eight SideStaking-managed BPool clones, and used asymmetric join and exit math to turn repeated single-sided liquidity operations into 127,861.011181 mOCEAN of net profit.
The attacker reused one non-owner signature across 41 permit calls on six Lixir vault clones, gained allowances over victim shares, and used withdrawETHFrom and withdrawFrom to extract USDC, USDT, ETH, and illiquid LIX from Uniswap V3-backed positions.
Introduction
There are already good writeups on CVE-2023-36802, a type confusion bug in Microsoft’s Streaming Service Proxy driver mskssrv.sys. Most of them were written before recent mitigations closed off the usual paths. This one was written after.
The bug was known. The patch was out. The writeups existed. My mentor Dang Nguyen (@MochiNishimiya) gave it to me anyway.
The constraint was simple: exploit it without relying on NtQuery* APIs1 for kernel address leaks or PreviousMode23 for arbitrary read/write. Build the exploit the way you’d have to build it today, under the mitigations that exist now, not the ones that existed when the bug was first disclosed.
DLMC counted attacker-funded USDT reserves as backing while excluding most freshly minted contract-held DLMC from the circulating-supply denominator, so two large `buy()` calls plus referral reward minting pushed `livePrice` upward before an immediate `sell()` extracted roughly $222.6K net.
TALOS-2026-2444
An infomation overwrite vulnerability exists in the AddGenericPassword functionality of Chrome 148.0.7778.216 (Mac arm64). A keychain write from a same-user process can overwrite Chrome’s encryption key, leading to disclosure of sensitive information. An attacker can make a...
TALOS-2026-2411
CVE-2026-12488
A memory corruption vulnerability exists in the GV-Cloud functionality of GV-VMS V20 (version(s): 20.0.2). A specially crafted network request can lead to a denial of service. An attacker can impersonate the legitimate server to trigger this vulnerability.
The...
TALOS-2026-2379
CVE-2026-12486,CVE-2026-12849,CVE-2026-12850,CVE-2026-12851
Multiple exploitable OS command injection vulnerabilities exist in the libNetSetObj.so functionality of GV-I/O Box 4E (version(s): 2.09). A specially crafted network packet can lead to command execution. An attacker can...
The attacker submitted 100 duplicate tier-42 LDA IDs with zero transfer amounts, creating a false 100x custom tier balance that Royalties trusted for claims and using a flash-swap-funded deposit to extract 263,808.953900 USDC before repayment.
Bridge delegated message authenticity to `SignalService.proveSignalReceived()` against a saved checkpoint, and the accepted checkpoint let forged source signals without legitimate Taiko `MessageSent` events mark messages valid or retriable before ERC20Vault and ETH releases.
TALOS-2026-2382
A denial of service vulnerability exists in the DNSSEC DNSKEY Extended Flags functionality of BIND (version(s): 9.21.21). A specially crafted mirror domain can lead to a denial of service. An attacker can serve a malicious zone to trigger this vulnerability.
The versions below...
The attacker used flash-borrowed capital to trade through the JB route contract, repeatedly triggering a privileged JB token path that burned tokens directly from the PancakeSwap V2 JB/USDT pair and called sync() after each sell to manipulate reserves.
Key Points Introduction In this research, we analyze a clipboard hijacker campaign that is hidden inside a collection of “solutions” and “tools” that claim to give users an unfair advantage. These offers include Solana and Pump.fun sniper bots (automated tools that try to buy new tokens or meme coins faster than other traders), Aviator Predictor […]
The attacker used flash liquidity to inflate the LBP/USDT pair reserve that LBPHashrate trusted during LP settlement, causing oversized hLBP credit, repeated LBP reward emissions, and a final exit of about 610.56 WBNB.
The attacker called the bridge's escape-hatch path with a proof whose public output named the attacker as recipient, causing the RollupProcessor to trust the forged output owner after verifier success and release 1158 ETH from bridge custody.
# Introducing Session Switcher. Swap Burp Sessions with One Click!
17 Jun 2026 - Posted by Savio Sisco
Authorization testing is one of the most repetitive, yet critical tasks in web app security testing. Checking for horizontal and vertical privilege escalation, IDORs, and other access control...
Microsoft Browser Vulnerability Research
Jun 16 2026 @ 7:00 AM
Microsoft Edge Extensions Security Team
119 extensions, up to 2.6 million installs, delayed execution, and payloads hidden inside image and font files. Here’s what we found, how we disrupted it, and how we’re protecting you.
AWS Forensics : What you need to know
