Cyber Threat Brief — 2026-02-28

Summary

2 new items met the 24-hour rule (both original disclosures on GitHub Advisories). Highest severity: Critical (Vikunja account takeover via password reset token reuse).

1. Vikunja Password Reset Token Reuse → Persistent Account Takeover (CVE-2026-28268)

What’s New

A logic flaw means password reset tokens remain valid indefinitely (not invalidated on use + broken cleanup cron), enabling persistent account takeover if a single token is exposed.

Technical Details

Field	Value
CVE	CVE-2026-28268
CVSS	9.8 (per GitHub Advisory page UI)
Affected	Vikunja API versions < 2.1.0
Exploit	No public PoC observed in GitHub PoC search (last 24h)
CISA KEV	No

Key mechanics (per advisory):

Reset flow deletes the wrong token kind (email confirm instead of password reset), leaving the reset token valid.
Cleanup job’s time comparison is inverted, deleting new tokens and keeping old ones.

TTPs

Tactic	Technique	Observable
Initial Access	T1190 — Exploit Public-Facing Application	Exploitation against exposed Vikunja API endpoints; unusual password reset activity from internet IPs
Credential Access / Persistence	T1078 — Valid Accounts	Successful logins from new IPs/ASNs after password reset; repeated resets over time using same token

Detection Opportunities

Focus on reset-token exposure + abnormal password resets:

Alert on high-rate password reset requests per user/IP.
Alert when password reset is followed by login from a new geo/ASN/device within a short window.
If you log request parameters: watch for repeated use of the same reset token (should never happen).

Sample detection (Sigma-style, web/app logs)

title: Vikunja Suspicious Password Reset Burst
id: 4e4cbd0d-34b5-4dbf-9d38-7ddc5c7c0d4d
status: experimental
description: Detects abnormal bursts of password reset requests that may indicate token abuse or takeover attempts.
author: AJ Threat Brief (generated)
logsource:
  category: webserver
  product: generic
  service: http
selection:
  cs-method: 'POST'
  cs-uri-stem|contains:
    - '/api/v1/user/password/reset'
condition: selection
fields:
  - src_ip
  - user
  - http_status
  - user_agent
falsepositives:
  - Legitimate user troubleshooting
level: medium

Log Sources

Reverse proxy / WAF (URI, IP, UA)
Application logs (token issuance/validation; user id)
Authentication logs (successful logins, session creation)

Detection Coverage

Source	Status
Sigma	❌ Gap (coverage-check returned 0 results)
Splunk ESCU	❌ Gap (coverage-check returned 0 results)
Elastic	❌ Gap (coverage-check returned 0 results)
KQL	❌ Gap (coverage-check returned 0 results)

Sources

GitHub Advisory Database (Vikunja) — https://github.com/advisories/GHSA-rfjg-6m84-crj2 — Published 2026-02-28

2. OpenClaw ACP Client Permission Auto-Approval Bypass via Untrusted Tool Metadata (GHSA-7jx5-9fjg-hp4m)

What’s New

A permission-bypass in the OpenClaw ACP client can allow auto-approving actions by abusing untrusted tool metadata, undermining human-in-the-loop controls.

Technical Details

Field	Value
CVE	N/A
CVSS	Medium (per advisory severity)
Affected	OpenClaw ACP client (see advisory for affected versions)
Exploit	Proof-of-concept not observed in GitHub PoC search (last 24h)
CISA KEV	N/A

TTPs

Tactic	Technique	Observable
Execution	T1059 — Command and Scripting Interpreter	Unexpected tool invocations / command executions initiated by the agent without explicit approval
Credential Access	T1552.001 — Unsecured Credentials: Credentials In Files	Agent attempts to read common secret files (env, config, key material) following metadata-driven tool calls

Detection Opportunities

Treat this like “agent control-plane abuse”:

Alert on tool calls that should require approval but occur without a corresponding approval event.
Alert on unusual file access patterns by the agent process (reading dotfiles, ssh keys, API tokens).

Sample detection (Sigma-style, endpoint telemetry)

title: Suspicious OpenClaw Agent Tool Execution Without Expected Approval Trail
id: 7ad0a6a1-7e45-4d9c-a1df-8eec2bba5d62
status: experimental
description: Flags unexpected executions from an OpenClaw agent process that may indicate control-plane abuse.
author: AJ Threat Brief (generated)
logsource:
  category: process_creation
  product: macos
selection:
  ParentImage|contains:
    - 'openclaw'
  Image|endswith:
    - 'node'
    - 'bash'
    - 'zsh'
condition: selection
fields:
  - ParentImage
  - Image
  - CommandLine
falsepositives:
  - Legitimate skill automation
level: medium

Log Sources

OpenClaw gateway / ACP client logs (approval events, tool call metadata)
Endpoint process creation telemetry
File access/audit logs where available

Detection Coverage

Source	Status
Sigma	❌ Gap (coverage-check returned 0 results)
Splunk ESCU	❌ Gap (coverage-check returned 0 results)
Elastic	❌ Gap (coverage-check returned 0 results)
KQL	❌ Gap (coverage-check returned 0 results)

Sources

GitHub Advisory Database (OpenClaw) — https://github.com/advisories/GHSA-7jx5-9fjg-hp4m — Published 2026-02-27

Intelligence Sources

Web search, X/Twitter presets, GitHub Advisories, CISA KEV