Beat
AI
How AI is changing attacks, defense and the security of the models themselves.
- 32
- stories
- OCT 7
- latest
Patching Faster Has Never Been the Answer
Patching is a necessary evil and doing it manually in any decent sized organization was essentially impossible in the pre-AI times and is now pure folly.
DENNIS FISHER · OCT 7 · 3 min
All AI
‘We Have Enough AI’: Realism in a Time of Relentless Hype
Let’s ask a simple question that seems to be getting lost in all of the credulous fervor around AI: How’s this working out for everyone?
Dennis Fisher · 3 min
Here Come the AI-Generated BEC Scams
Messages delivered through legitimate third-party email infrastructure had C-suite sender display names, custom signatures, and direct approval notes.
Dennis Fisher · 2 min
One Year of Decipher (Relaunched)
It’s been one year since we relaunched Decipher in September 2025. Here are the videos, podcasts, and articles that best reflect our mission statement: Security without fear.
Decipher · 4 min
Linux Foundation Backs New TRACE AI Security Standard
TRACE aims to help prove that sensitive data in organizations is being handled “according to policy” by AI agents and open weight models.
Lindsey O'Donnell-Welch · 2 min
Inside ExploitGym: How Researchers Are Measuring AI Agent Exploitation Capabilities
A research collaboration, directed by UC Berkeley, led to the development of ExploitGym, a large-scale benchmark for measuring AI agents’ abilities to develop exploits. Here’s the story of how ExploitGym came to be.
Lindsey O'Donnell-Welch · 9 min
MLflow Bug Actively Exploited to Steal Credentials
This flaw allows remote, unauthenticated attackers to read sensitive data from internal network services or cloud metadata endpoints.
Dennis Fisher · 2 min
OpenAI, Anthropic AI Agents Performed ‘Unsanctioned’ Actions During Cyber Tests
OpenAI and Anthropic models made “unsanctioned” actions during tests by the AI Security Institute. Here’s what we know.
Lindsey O'Donnell-Welch · 5 min
OpenAI-Hugging Face Breach: Five New Things We Learned
New details emerged this week about an intrusion by OpenAI models on Hugging Face’s infrastructure.
Lindsey O'Donnell-Welch · 4 minYour LLM Is Showing: New Data Finds Sharp Rise in Exposed AI Tools
Exposures of AI/LLM tools has risen by more than 60 percent over the last nine months.
Dennis Fisher · 2 min
Bug in Cursor Allows Simple RCE
This exploit occurs entirely in the background, requiring no user approval, dialogue prompts, or interaction.
Dennis Fisher · 3 min
New Forg365 PhaaS Kit Emerges
Like EvilTokens and Kali365, Forg365 is distributed through Telegram and offered as a commoditized service for attackers who want to target Microsoft 365 environments.
Lindsey O'Donnell-Welch · 2 min
Researchers Find New GhostApproval Bug in Many AI Coding Assistants
The bug is a modern twist on a Clinton-era piece of known Unix badness: following symbolic links.
Dennis Fisher · 4 min
macOS Gaslight Backdoor Weaponizes Prompt Injection Against Security Analysts
Gaslight is a Rust-based backdoor that researchers from SentinelOne have attributed to DPRK-aligned threat actors.
Dennis Fisher · 3 min
Mastra AI Hit By npm Compromise
The unknown threat actor gained unauthorized access to legitimate npm accounts, allowing them to inject malicious dependencies into widely used packages.
Dennis Fisher · 2 min
Anthropic Warns of LLMs’ Impact on (Already Shrinking) N-Day Exploit Gap
Anthropic researchers warned that modern AI tools are further intensifying the existing issue of threat actors creating N-day exploits.
Lindsey O'Donnell-Welch · 4 min
A Copy Fail FAQ
The latest branded bug is a slick, portable LPE in many Linux kernels going back to 2017.
Dennis Fisher · 3 min
Vercel Says Internal Systems Hit in Breach
The incident came to light on Sunday and the company says it has brought in an incident response provider to investigate the intrusion. Details of the intrusion are scant at this point.
Dennis Fisher · 3 min
