Claude Code settings files expose secrets on public GitHub
About 1,100 of more than 100,000 uploaded files, from 63 organizations, held API tokens, passwords, or private keys.
1
Stories per day · 30 days
4 RSS
Relatedai-agents3anthropic3secrets3agent-security2api-keys2claude2claude-code2environment-variables2fortigate2fortios2opencode2railway2accenture1active-directory1advisory1agentic-ai1ai-attack1amap1
26 stories, newest first · one card per story
Stories tagged #credential-exposure
Chainalysis Breach Exposes All Users' Credentials
AI-driven on-path attack exposed names, usernames, password hashes and live tokens for all Chainalysis users; KYT and login page taken offline.
1
Northflank security breach leads to unauthorized AI API key usage
Northflank reported possible exposure of environment variables containing AI API keys after a Wiz threat alert, urging customers to rotate secrets.
1
Bishop Fox recovers Claude sandbox tokens and bypasses a local block
Anthropic said the observations stay inside the untrusted sandbox and do not qualify as security issues.
Accenture employee allegedly gave Claude Code admin access to Pentagon systems
WIRED says the agent ran on unclassified NIPRNet for about two months and may have sent credentials off-network; Accenture calls the claims meritless.
1
ThreatsDay: Fortinet fixes, PHP2Shell scans, Authenticator token flaw
Fortinet is withholding flaw details until November 10 while Hudson Rock reports 40,573 exposed FortiGate secrets.
.png&w=828&q=75)
26,400 live credentials found scanning 7.25M Gitee repos
TruffleHog and Kingfisher verified cloud keys and an unnamed enterprise AI platform's organization-admin credential.
4
FortiBleed 2.0 finds 40,573 cleartext secrets in FortiGate configs
Hudson Rock parsed 1,703 FortiGate configuration reports and 1,689 host identifiers; 975 domain labels were absent from the earlier set.
FortiBleed 2.0 finds 40,573 cleartext secrets in exposed FortiGate configs
The follow-up is stolen configuration files, not a new Fortinet flaw, and the records are not confirmed still valid.
Railway warns developers to rotate secrets after possible breach
An unauthorized party may have decrypted some environment variables; deployments stay paused while builds return in stages.
Malware planted on Onamae.com Desktop Cloud; 26,739 servers flagged
GMO shut every Desktop Cloud plan after October 20, and OANDA warned MT4 and MT5 logins may have been viewed.
Railway says intruders may have read customer environment variables
Possible secrets include API keys and registry credentials; builds were still paused at 02:19 UTC on October 22.
1
OpenCode Console breach may have exposed provider API keys
OpenCode says 1.4% of Console users with linked external providers were affected; the OpenCode product itself was not.
OpenCode breach may have exposed Console users’ API keys
OpenCode says 1.4% of Console users who linked external providers were emailed on October 17.
CERT-In Warns Hackers Using AI Agents to Launch Cyberattacks in Just Hours
CERT-In says attackers and runaway AI agents reached Indian systems and data within hours.
AI Agent Workflows Could Expose Credentials Through Trusted Developer Tools
Trellix warns AI agent workflows can leak credentials from vaults into commands, logs, and model context.
6
Researchers tracked down a rogue fleet of Chinese AI agents in the wild and actually talked to it
Researchers found Tencent Cloud AI agents scraping Alibaba's Amap, bypassing bot blocks, and replied to hidden instructions.
2
Amazon Bedrock AgentCore Flaws Could Expose AWS Credentials
Two Amazon Bedrock AgentCore SDK flaws (CVE-2026-12530, CVE-2026-16796) enabled command execution inside AI sandboxes and exposure of attached AWS credentials; patched in 1.18.1.
3
CVE-2026-81930: Apache Airflow Snowflake provider: Unvalidated account field redirects SQL API bearer token off-domain
Apache Airflow's Snowflake provider can leak SQL API bearer tokens off-domain via an unvalidated account field (CVE-2026-81930).
Storm-3168: Agentic-driven cloud attacks using compromised service principals
Microsoft details Storm-3168 (JADEPUFFER) destroying Azure resources via compromised service principals in the first documented agentic ransomware operation's Azure activity.
1
AWS Automatically Quarantines Exposed IAM Keys Within 10 Seconds of GitHub Leak
AWS quarantined a leaked IAM key 10 seconds after Unit 42 published it to public GitHub.
Jenkins Patches 20 Plugin Flaws Leading to RCE, XSS and Credential Theft
Jenkins patched 20 vulnerabilities across 13 plugins, including Groovy sandbox bypasses enabling remote code execution on CI/CD controllers.
3
Surfshark VPN says hackers breached internal testing, proxy servers
Surfshark disclosed that hackers accessed misconfigured internal test and proxy servers, exposing build credentials but not customer data, VPN traffic, or production infrastructure.
2
CVE-2026-49364: Apache Artemis, Apache ActiveMQ Artemis: Pre-Authentication Cluster Credential Exposure to Discovered Peers
Apache Artemis CVE-2026-49364 exposes cluster credentials to network-adjacent peers discovered before authentication completes.