Attackers weaponise SAP flaws within days as AI bills soar

SecurityWeek’s latest weekly briefing notes several high‑profile developments across device hardware, enterprise AI abuse, and software vulnerabilities. Among the items, a buyer reported an implant inside a Trezor Safe 5 unit bought from CryptoBilis, with Trezor advising customers to move funds to a newly created wallet. The firm has not yet publicly verified the claim; the reseller previously linked to tampering of Ledger devices.

Separately, Mandiant’s investigations into LLMJacking describe attackers monetising compromised AI model access, generating fraudulent bills up to US$1 million in total for some organisations, with initial access often via stolen credentials and API keys, and a LiteLLM vulnerability (CVE-2026-40217) used to elevate permissions in one incident. Google Threat Intelligence Group also flagged LLM Mixers, including Jet Night, as marketplaces for stolen keys.

On the vulnerability front, Onapsis and Recorded Future report that SAP flaws patched in 2025–2026 now show a dramatically shorter exploitation window, with a median of one to three days between a fix and first exploit or PoC, and ransomware operators increasing interest in SAP. Synology warns that attacks are targeting devices running outdated DSM versions, exploiting a vulnerability fixed in December 2024 (Synology‑SA‑24:20), with automatic updates recommended.

Additional items include Flowise “hello” attacks abusing CVE-2024-58351 to execute commands and gather credentials, and a broader note that the Chainalysis breach expanded to expose user names, password hashes and live tokens, prompting offline services and customer outreach.