ai-security · 7 min
On 27 July, Zou et al. publish a paper showing that adversarial suffixes can be generated automatically, by gradient descent, that bypass the safety classifiers of Llama-2, GPT-3.5, GPT-4, Bard and Claude. And they transfer between models.
· Manuel López Pérez
news · 4 min
MOVEit changes everything: Cl0p turns a SQLi zero-day into a campaign that ends with 2,700+ affected organisations. Meanwhile, Fortinet, VMware Aria, and the first LLM agents in production.
· Manuel López Pérez
tutorials · 8 min
CVE-2023-34362 is a pre-auth SQL injection in MOVEit Transfer that Cl0p exploits as zero-day from 27 May. The chain goes SQLi → MachineKey leak → session forge → LEMURLOOT web shell drop (human2.aspx). Result: 2,700+ organisations exposed before the year ends.
· Manuel López Pérez
news · 4 min
Barracuda ESG compromised for 7 months by UNC4841, Microsoft Build sells Copilot for everything, Operation Triangulation reveals an iOS implant. May brings geopolitics in the background and AI security turns into product marketing.
· Manuel López Pérez
tutorials · 8 min
CVE-2023-2868 is a trivial command injection in the Email Security Gateway TAR parser. UNC4841 has been exploiting it since October 2022. Barracuda does not recommend patching — it recommends throwing the device out. We reproduce the malicious filename, walk through the buggy Perl path and review the IoCs published by Mandiant.
· Manuel López Pérez
news · 7 min
Markdown exfil moves out of the forum and into serious advisories. MSI Gaming leaks Intel Boot Guard signing keys. QueueJumper in MSMQ. Anthropic opens the Claude API. Meanwhile, Cl0p keeps adding victims.
· Manuel López Pérez
ai-security · 6 min
A chatbot that renders markdown turns any `` into an automatic GET to that URL. If an attacker can inject markdown via indirect prompt injection, they exfiltrate the entire context. Reproducible PoC.
· Manuel López Pérez
news · 4 min
Outlook NTLM zero-click patched after a year of APT28 exploitation, 3CX supply-chain compromise with Lazarus DLL sideloading, GPT-4 released and broken within 24 hours. March closes Q1 with two long stories.
· Manuel López Pérez
tutorials · 8 min
CVE-2023-23397 lets Outlook send the user NTLMv2 hash to an attacker SMB server just by processing an email. No click, no preview. Exploited as zero-day by APT28 since April 2022. Reproducible PoC with Outlook COM, Wireshark capture of the handshake and offline crack with hashcat.
· Manuel López Pérez
news · 5 min
Sydney leaked, Greshake formalises indirect prompt injection, Cl0p turns GoAnywhere into a global campaign, and ESXiArgs sweeps unpatched VMware ESXi. February opens two fronts at once.
· Manuel López Pérez
ai-security · 8 min
On 8 February Kevin Liu pulled the system prompt out of Bing Chat with an "ignore previous instructions". On 23 February Greshake published the paper that defines the next wave: instructions embedded inside the content the LLM reads.
· Manuel López Pérez
news · 6 min
BOLDMOVE on FortiOS attributed to a China-nexus actor, ALPC zero-day in Patch Tuesday, MFA seeds stolen from GoTo, NIS2 enters into force, and the first generation of ChatGPT jailbreaks. What moved in January, ranked by real impact.
· Manuel López Pérez
