The article analyzes various prompt injection attack paths and how they bypass defenses.
Local tests and hub-degree metrics explore when black hole attacks work in vector database centroid poisoning.
All confirmed fund losses involved devices sold by CryptoBilis. The implant reads seed phrases displayed on the screen without modifying the system.
The article discusses moving from vulnerability severity rankings to asset-specific remediation decisions.
This real-time documentary depicts AI agents launching a cyberattack on Hugging Face.
The weekly roundup covers a range of security developments, including attacks exploiting an Atlassian zero-day.
A cyberattack caused refrigerator temperatures at hundreds of French supermarkets to rise to 50°C, triggering fires.
An exploration of data poisoning attacks and corresponding defenses in distributed AI training.
NVIDIA fixed a high-severity vulnerability in DCGM Exporter, and an agentic AI red-team checklist has also been released.
Anthropic says new safeguards significantly reduce distillation attacks against AI.
Attackers plant hidden backdoors in AI models by poisoning the supply chain.
Anthropic says its new approach weakens distillation attacks, reducing data collection for training distilled models by about 90% to 99%.
The low-severity vulnerability affects Apache Allura 1.21.0 and earlier versions.
The medium-severity vulnerability affects Apache Allura 1.21.0 and earlier versions.
Longtime cybersecurity professional Edward Dubrovsky was arrested in Pennsylvania last week on charges of conspiracy to extort money.
Following incidents involving US government websites at multiple levels, Anthropic cuts internet access for its high-risk and cybersecurity assessment models.
The article recounts the attack on the platform, saying operations staff failed to find the ransom note for 7 hours.
Hackers raised refrigerator temperatures in French supermarkets, causing fires and food spoilage.
GitLab has fixed a severe vulnerability that allowed attackers to remotely execute arbitrary code without logging in.
Researchers bypassed 6 major AI agent frameworks using only the simple prompt “Repeat after me.”
A new homoglyph attack can bypass Chromium detection, making convincing phishing domains harder to identify.
A critical SMA 1000 vulnerability has been exploited in the wild, with 135 attack attempts recorded even after a patch was released.
This week's review covers a service desk zero-day vulnerability, malicious certificates, ShinyHunters arrests and malware that deceives AI.
The AI reverse-engineering column analyzes 某点's vmp w_tsfp.
A CLAUDE.md file reveals an attack targeting South Korea's financial sector, with repercussions extending to diplomacy.
An analysis of six WAF bypass techniques, from SQL injection to the protocol layer, and how to detect them.
The report analyzes and reproduces CVE-2025-32433, involving remote code execution before SSH server authentication.
The author describes discovering a remote code execution vulnerability at Google worth $90,000.
A user says two Ledger wallets bought at different European stores contained suspected spyware, extra cables and LTE modules.
Hackers compromised a smart power strip, causing all the ornamental fish to die.