From Untrusted Input to Trusted Memory: A Systematic Study of Memory Poisoning Attacks in LLM Agents

Memory is a core component of AI agents, enabling them to accumulate knowledge across interactions and improve performance. However, persistent memory introduces the risk of memory poisoning, where a single adversarial memory write can exert long-term influence over agent behavior. We present a...

PT-2026-45877

Name of the Vulnerable Software and Affected Versions openSeaChest version 25.05.3 Description An out-of-bounds write exists in the --showSupportedFormats command. This occurs when a maliciously crafted NVMe device provides a bogus value in the namespace FLBAS byte, allowing one extra byte to be...

PT-2026-46566

Name of the Vulnerable Software and Affected Versions Google Chrome versions prior to 149.0.7827.53 Description An out of bounds write occurs in Codecs, which could allow a remote attacker to potentially perform a sandbox escape by using a crafted video file. A sandbox escape is a technique used ...

PT-2026-45745

Path traversal in restore handler in Collibra Agent, allows an attacker to write arbitrary files via a crafted ZIP archive. Collibra Agent fails to properly validate and canonicalize file path during ZIP extraction, this can allow an attacker to write files outside the intended extraction directo...

PT-2026-45750

Name of the Vulnerable Software and Affected Versions Gleam versions 1.16.0 through 1.17.0 Description A path traversal issue exists in the handling of custom documentation pages. The documentation.pages entries within the gleam.toml file are incorporated into filesystem paths without sufficient...

SUSE SLES15 Security Update : kernel RT (Live Patch 10 for SUSE Linux Enterprise 15 SP7) (SUSE-SU-2026:2131-1)

The remote SUSE Linux SLES15 host has a package installed that is affected by multiple vulnerabilities as referenced in the SUSE-SU-2026:2131-1 advisory. This update for the SUSE Linux Enterprise Kernel 6.4.0-150700.7.34 fixes various security issues The following security issues were fixed: -...

CVE-2026-45691

A flaw was found in Nextcloud Server. An attacker could reuse a pre-two-factor authentication 2FA session cookie as a Bearer token. This allows them to authenticate against DAV endpoints, granting unauthorized read and write access and bypassing the mandatory two-factor authentication. Mitigation...

CVE-2026-25259 Out-of-bounds Write in DSP Service

Memory corruption while processing multiple IOCTL command for escape operations...

CVE-2026-24088 Missing Authentication for Critical Function in Boot

Cryptographic Issue while processing a specific partition which allows unauthorized write access to load a customized bootloader...

CVE-2026-24088

Technical details are not publicly available in the provided documents for CVE-2026-24088. Monitor for updates.

CVE-2026-24088 Missing Authentication for Critical Function in Boot

Cryptographic Issue while processing a specific partition which allows unauthorized write access to load a customized bootloader...

CVE-2025-59614 Out-of-bounds Write in Windows Compute

Memory Corruption when sending random number generator command with insufficient output buffer size...

CVE-2025-59611 Out-of-bounds Write in Core Services

Memory corruption in diagnostic services due to absence of input validation...

CVE-2025-59606

The CVE-2025-59606 entry describes a memory corruption flaw triggered by writing to invalid memory locations caused by heap exhaustion during secure data initialization. The CVSS 3.1 vector indicates a local, low-privilege, no-user-interaction exposure with high impact to confidentiality, integri...

CVE-2025-59605 Out-of-bounds Write in HLOS

Memory Corruption when processing device identifier strings that exceed the expected maximum length...

CVE-2025-59605 Out-of-bounds Write in HLOS

Memory Corruption when processing device identifier strings that exceed the expected maximum length...

CVE-2026-0100

In Load of LoadedArsc.cpp, there is a possible out of bounds write due to a heap buffer overflow. This could lead to local escalation of privilege with no additional execution privileges needed. User interaction is not needed for exploitation...

CVE-2026-10279

A vulnerability was identified in hiraishikentaro wezterm-mcp 0.1.0. The affected element is an unknown function of the file src/weztermexecutor.ts of the component switchpane/writetospecificpane. The manipulation of the argument request.params.arguments.paneid leads to os command injection. The...

CVE-2026-43624

F5-TTS up to v1.1.20 contains a path traversal vulnerability in the finetune Gradio handlers that allows unauthenticated attackers to write arbitrary files by passing unsanitized project names to os.path.join() without validating the resulting path. An attacker can supply absolute paths (e.g., /t...

CVE-2026-43624

F5-TTS through version 1.1.20 contains a path traversal vulnerability in the finetune Gradio handlers that allows unauthenticated attackers to write arbitrary files by passing unsanitized user-supplied project names directly to os.path.join without validating the resulting path stays within the...