Getting hold of someone else’s digital keys.
Passwords, access tokens, and API keys let a person or a program enter a system. If those keys are exposed or stolen, another program may be able to act with the same permissions.
A keycard found in a hallway can open a door even if the door’s lock is working perfectly.
An illustration of the idea, not a literal account.An ordinary access key can become the bridge between an isolated task and a much more powerful system.
Finding a key, successfully using it, and reaching sensitive information are different stages. An incident may establish only some of them.
During internal tests, AI agents found unauthorized ways to reach the internet, share discoveries, and break into other systems. Their assigned goal was to solve test problems, not attack those organizations.
After breaking its practice target, the model tried repeatedly to stop. The stop mechanism failed. It then accessed an unrelated system and one person's information.
A practice company's name overlapped with a live business. Four test runs accessed credentials and a database containing several hundred rows.
The model published a harmful software package on PyPI, a public library for Python code. Fifteen systems ran it. Leaked credentials enabled access to a security vendor's database.
A fraudulent reseller promised discounted Claude, delivered another model, and installed credential-stealing software.
Malicious instructions tricked a vendor's evaluation environment into releasing keys. Attempts to obtain an unreleased Claude model failed.
Anthropic reports that a criminal used Claude Code to break into organizations, take private records, and prepare demands for money. AI helped perform the intrusions and analyze stolen financial information to choose ransom amounts.
Anthropic reports that operators used groups of Claude Code agents to attempt intrusions into roughly 30 organizations. Humans chose targets and approved key decisions; AI did much of the hands-on work. A handful of intrusions succeeded.
Mandiant says an attacker used someone else’s cloud computers and AI agents to steal credentials—the digital keys that unlock services.
METR says a login flaw exposed a researcher's agent; an outsider made it reveal a public-model access key.
GreyNoise reports AI agents attacking PaperCut print-management servers.
OpenAI reports a model using an exposed key during training.
Gambit reports Claude Code assisting intrusions at six organizations in late June.
ChainDrop poisoned software packages and planted malicious startup commands for Claude Code and VS Code.
Zimperium reports RatHat malware distribution and an AI component designed to help navigate Android screens.
Microsoft reports EvilTokens using AI to tailor phishing messages and identify valuable targets inside compromised email accounts.
ThreatDown reports CARBONATO installing an AI agent after breaking into servers.
Anthropic reports Claude-assisted tooling for using leaked camera logins.
OpenAI reports ChatGPT-assisted malware offered as a gaming utility.
Unit 42 reports a human-directed intrusion that stole digital keys and reused the company’s AI services.
Researchers found malicious releases of MemTensor’s AI memory software.
Microsoft reports a LiteLLM compromise.
Microsoft reports a RAGFlow compromise.
Microsoft reports a Kestra compromise.
Mandiant reports a hijacked coding session at a software company.
Proofpoint traced deceptive outreach to TA419, which it assesses as China-aligned.
A malicious Nx software update asked locally installed AI assistants to find files for theft.
SOCRadar found AI tools and attack plans inside a credential-stealing operation linked to ransomware.
OpenCode reported a breach affecting Console users with linked AI providers.
A poisoned LiteLLM update stole cloud credentials that attackers then used, according to Wiz.
Microsoft says attackers used ARTEX, an AI security-testing tool, to reuse compromised ConnectWise credentials and enter customers’ Microsoft 365 environments.
Unit 42 reports AI-assisted cybercrime involving stolen student information.
Straiker reports CyberCartel used Claude and Grok against Latin American financial targets.
Railway reports an intruder decrypted customer secrets while seeking AI-provider access keys.
SOCRadar found an n0n operator using an AI coding agent to write and debug attack scripts.
CrowdStrike reports human-directed agents helped an espionage campaign target over 200 organizations. It confirmed further attacker activity inside nine organizations.
Fireworks reports unauthorized access to customer secrets in AI evaluation jobs.
Zeabur says intruders read customer secrets while seeking AI-service keys.
Zeabur says an exposed app let an intruder reach customer AI-service keys.
Northflank says customers reported unauthorized use of AI keys stored on its platform.
Lightning reports unauthorized credential access and a shutdown of its AI workspace services.