Loading story…
CodeWhale version 0.8.64 fixes multiple high-severity vulnerabilities affecting the codewhale and codewhale-tui packages, including remote code execution, SSRF bypass, path traversal, and argument injection bugs. The most severe issue, CVE-2026-75858, allowed model-supplied Python code in rlm_eval to run automatically without honoring user approval settings, enabling local code execution through prompt-injection content delivered via web pages, fetched URLs, repository files, or MCP tool results. Additional flaws included CVE-2026-75856, a DNS pinning TOCTOU weakness that could bypass SSRF protections and reach internal network resources, and CVE-2026-75914, which let attackers abuse symlinks in image_analyze to read files outside the workspace and leak their contents to the vision endpoint.
The release also addressed Git tool injection risks that could expose or modify local files. CVE-2026-75912 allowed attackers to pass crafted rev values to git_blame and read arbitrary files through injected Git options, while CVE-2026-75913 let malicious input reach git_show without proper validation or an end-of-options guard, enabling arbitrary file writes to targets such as ~/.ssh/authorized_keys or shell configuration files. A related hardening commit shows CodeWhale tightened local tool trust boundaries by resolving hosts before fetch_url requests, sanitizing the child environment for JavaScript execution so parent secrets are not inherited, and canonicalizing image paths to block symlink escapes from the workspace.
Track how attackers are adapting to this technology.
7 events from the most recent confirmed update back to the earliest known activity.
On August 18, 2026, CVE-2026-75914 was published for a path traversal vulnerability in CodeWhale's image_analyze tool affecting versions before 0.8.64. The flaw let attackers use workspace symlinks to external files to leak file bytes to the vision endpoint, and the remediation was to upgrade to 0.8.64 or later.
On August 18, 2026, CVE-2026-75913 was published for an argument injection vulnerability in CodeWhale's git_show tool affecting versions 0.8.41 through 0.8.63. The flaw allowed attacker-controlled rev values to be interpreted as Git flags and could enable arbitrary file writes; version 0.8.64 fixed the issue.
On August 18, 2026, CVE-2026-75912 was published for an argument injection flaw in CodeWhale's git_blame tool affecting versions before 0.8.64. The issue allowed attacker-controlled rev values such as --contents=/path/to/file to read arbitrary files through tool output, with 0.8.64 listed as the fixed version.
On August 18, 2026, CVE-2026-75858 was published for a remote code execution issue in CodeWhale's rlm_eval tool affecting versions 0.8.41 through 0.8.63. The flaw let model-supplied Python execute without honoring the user's approval policy, and version 0.8.64 was identified as the fix.
On August 18, 2026, CVE-2026-75856 was published for a DNS pinning TOCTOU flaw in CodeWhale before 0.8.64 that could let remote attackers bypass SSRF protections and reach internal IP addresses. The advisory identified version 0.8.64 as the fix and referenced commit 26de44a.
On June 21, 2026, Hmbown committed change 4356335 to the CodeWhale repository to stop project-local TUI configuration from loosening security policy. The change prevents project config from enabling shell access or replacing user-owned instruction file lists, while still allowing project config to disable shell access and emitting warnings for ignored settings.
On June 21, 2026, Hmbown committed change 26de44a to the CodeWhale repository. The commit added hostname preflight checks for fetch_url, sanitized the child environment for js_execution, and enforced canonicalized workspace-bound path validation for image_analyze, with tests covering each change.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
Follow how adversaries are adapting to this technology, and where it touches your stack today.
9 references tracked. Mallory keeps watching after this page renders.
cvefeed.io
Open sourcecvefeed.io
Open sourcecvefeed.io
Open sourcecvefeed.io
Open sourcecvefeed.io
Open sourcevulncheck.com
Open sourcegithub.com
Open sourcegithub.com
Open sourcegithub.com
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
A high-severity path traversal vulnerability tracked as **`CVE-2026-55629`** was disclosed in **Whistle**, the HTTP, HTTP2, HTTPS, and WebSocket debugging proxy. The flaw affects versions prior to **`2.10.3`** and stems from the **`/cgi-bin/temp/get`** handler in `lib/service/service.js`, which passed a user-controlled filename directly to file retrieval logic. An unauthenticated remote attacker could exploit the issue without user interaction to read arbitrary files on the host, including sensitive system files such as `/etc/passwd`. The issue received a **CVSS 4.0 score of 8.7** and was addressed in Whistle **`2.10.3`**, with public references pointing to the project release, a fixing commit, and a GitHub security advisory. Subsequent project changes also show broader hardening in Whistle's service and temporary-file handling, including centralized temp-file utilities and a UID-based authorization check for service endpoints, alongside other fixes to proxy authorization, parsing, and plugin-related logic.
Jul 17, 2026
Warp disclosed and patched multiple command injection vulnerabilities that could let attacker-controlled input reach shell commands in the terminal and development environment. One flaw, tracked as **CVE-2026-48732** / `GHSA-qqpc-wvvw-4269`, affected legacy SSH background command handling from version `v0.2023.03.21.08.02.stable_00` onward. Malicious remote host, repository, or directory names could inject shell syntax into helper commands used for SSH-backed metadata collection, allowing command execution on the remote host with the victim’s authenticated SSH account after user interaction in an affected legacy SSH session. Warp said the issue was fixed by escaping embedded single quotes in the remote working directory and released patched versions including `v0.2026.05.06.15.42.stable_01` and `v0.2026.05.13.09.15.stable_01`. A separate high-severity flaw, **CVE-2026-48719**, affected Warp versions from `0.2025.08.06.08.12.stable_00` through `0.2026.05.06.15.42.stable_01` and allowed a malicious Git branch name to be interpreted by the victim’s shell when selected from Warp’s prompt branch selector. Warp’s June code changes also show broader hardening of remote SSH session handling, including shell-specific escaping for single quotes before constructing commands such as `cd` and `cat`, plus tests covering Bash, Zsh, Fish, and PowerShell to keep crafted paths inert. The combined disclosures indicate that untrusted repository metadata and remote session context could be weaponized for command execution unless users upgrade to fixed releases.
Jun 29, 2026
Wireshark disclosed and patched a broad set of vulnerabilities across its packet analysis engine, protocol dissectors, codecs, decompression routines, and the `sharkd` backend, with the most serious issues including possible code execution, memory corruption, and denial of service. The release notes for **Wireshark 4.6.6** aggregate advisories `wnpa-sec-2026-08` through `wnpa-sec-2026-50`, while earlier coordinated fixes shipped in **4.6.5** and **4.4.15** for affected `4.6.0-4.6.4` and `4.4.0-4.4.14` branches. One high-severity flaw, tracked as **CVE-2026-5656**, allowed a crafted configuration-profile ZIP to escape the intended extraction path and plant a Lua plugin for execution on the next launch. Additional issues included crashes and memory-safety bugs in AMR-NB (`CVE-2026-5654`), iLBC (`CVE-2026-5657`), ASN.1 PER (`CVE-2026-6527`), DCP-ETSI (`CVE-2026-5653`, `CVE-2026-6530`), and later `sharkd` and Bluetooth AVRCP flaws that could crash the process or potentially enable control-flow hijack. Several vulnerabilities were reachable simply by opening a malicious packet capture or by parsing malformed traffic on the wire. Wireshark documented infinite-loop and CPU-exhaustion conditions in the USB HID dissector (**CVE-2026-6534**), OpenFlow v5 dissector (**CVE-2026-6521**), and LZ77 decompression path (**CVE-2026-6533**), alongside crashes in BEEP (**CVE-2026-6538**) and other protocol handlers. GitLab issue reports show researchers and fuzzing jobs repeatedly triggering heap overflows, double-free conditions, NULL dereferences, and non-advancing-offset parser failures in components such as DCP-ETSI, H.245, MIH, eDonkey, and `sharkd`, underscoring that malformed captures can destabilize both the GUI and `tshark`. Wireshark said it was not aware of active exploitation at disclosure time and urged users to update to the fixed releases.
Aug 21, 2026