Sysdig's Threat Research Team documented an operator named JADEPUFFER that exploited CVE-2025-3248, a critical unauthenticated remote code execution flaw in Langflow patched in April 2025 and in CISA's KEV catalog since May 2025, to gain code execution on an internet-facing AI orchestration server. From there, an LLM agent conducted the entire operation autonomously: system reconnaissance, environment variable scanning for API keys and cloud credentials, a MinIO object store enumeration using default credentials that had never been changed, credential harvesting, persistence via a 30-minute crontab beacon, and pivot to a separate production MySQL and Nacos server.

The agent exploited CVE-2021-29441 to forge Nacos administrator accounts, encrypted 1,342 Nacos configuration items using MySQL's native AES_ENCRYPT function, dropped the original tables, and created a README_RANSOM table containing a Bitcoin address and a contact email. The AES key was generated randomly, printed to stdout once, and never stored or transmitted. The data is permanently unrecoverable regardless of payment. The agent self-corrected in real time: in one sequence it went from a failed login to a working fix in 31 seconds. Its payloads contained natural language annotations explaining each step's reasoning, which Sysdig used to confirm LLM authorship.

Huntress documented an automated password spray campaign running from June 12 to June 26 that generated more than 81 million authentication attempts against Microsoft 365 environments via the Azure CLI. The attacker used previously breached but never-rotated credentials from combo lists and validated them using the OAuth Resource Owner Password Credentials flow, a deprecated mechanism that sends a username and password directly to the /token endpoint and mints an access token without triggering an interactive MFA prompt. The campaign compromised 78 accounts across 64 organizations. Activity ceased after hosting provider LSHIY LLC terminated the accounts responsible for the IPv6 range in use.

Of the 23 businesses compromised during the June 22 spike, 15 had MFA implemented and enforced via Conditional Access Policy. Their MFA failed to fire because policies were scoped too narrowly: covering specific apps rather than all cloud apps, covering specific user groups rather than all users, or enforcing MFA only from untrusted locations. Eight impacted businesses had no MFA policy at all. Huntress reports that password spray volume across its customer base has increased by more than 155 times over the past six months, averaging 1,964 failed attempts per tenant per month.

DHS confirmed on July 2 that unknown threat actors breached the Homeland Security Information Network and an associated SharePoint collaboration system sometime between late May and early June 2026. HSIN is the federal government's primary platform for sharing sensitive but unclassified threat intelligence across federal, state, local, tribal, territorial, international, and private-sector partners. Attackers accessed HSIN servers and the SharePoint system before being detected. DHS isolated affected systems, launched a forensic investigation, and confirmed that classified networks were not impacted. Whether documents were exfiltrated remains under investigation.

The breach occurred during the run-up to the 2026 FIFA World Cup, which HSIN was being used to coordinate security planning across sixteen U.S. host cities. Senator Mark Warner called for DOJ and DHS investigations, saying the information in HSIN, while unclassified, is highly sensitive and its exposure risks national security. DHS has not publicly attributed the breach to any specific threat actor or nation-state.