AI Safety & Ethics

A few weeks ago, I started looking more seriously at AI “skills.” Not skills in the human sense. I...

Is Your AI Deepfake a Little Too Perfect? Detecting the Absence of Humanity ...

I’m building a small public-safe prototype called WorkflowGuard AI. The idea is simple: Before a...

Keyword blocking sounds like a simple fix. Block risky words. Stop unsafe prompts. Prevent sensitive...

The story matters, but maybe not for the reason the headline wants you to think "First...

The European Commission's new AI cybersecurity action plan promises structured access to frontier models. What it actually exposes is how little leverage Brusse

Part 1 of "Trust the Machine" -> a series on building AI infrastructure that is secure, compliant,...

In a previous post, I covered the one-pixel attack, where differential evolution finds a single pixel...

For years, the cybersecurity conversation about AI was speculative. Researchers warned it could lower...

A single crafted email made Microsoft 365 Copilot exfiltrate internal data with no user click. Here's the mechanism, and how to bound the blast radius when inje

Picture a mineral processing plant at 3 AM. A methane sensor in the crusher hall starts climbing....

Most AI incidents don't happen because the model gave a bad answer. They happen because nobody was...

The capability phase is over For the past two years, the AI conversation has been about...

I have released five short papers on one idea: making AI-assisted work re-checkable. All five are...

A language model that will not write a phishing email will still forward the confidential file, if a...

An AIBOM is not an SBOM with a model row bolted on. Here's what actually belongs in one, and why the vendor-supplied version won't save you.

Introduction AI agents are shipping fast — and so are the attacks against them. Prompt...

Recientemente, el análisis de vulnerabilidades en endpoints de IA (como se detalla en el artículo...

Every SAST scanner finds the obvious pattern: a tainted variable fed directly into an LLM call in the...

An AI finding you cannot reproduce is a liability, not a result. Darkmoon attaches the exact commands...

When I was building security auditing tools like Git Secret Scanner, the rules were binary: a...

TL;DR— Most production AI teams build safety layers using the content-moderation mental model:...

Governance is not a set of rules layered on top of AI. It is the structural logic that determines how...

Every aligned model pays the alignment tax. RLHF and DPO both trade reasoning capability for corporate agreeableness. Here is what that costs you — and why hone

Every "AI security analyst" I tried had the same flaw: a correct verdict and a confident-but-wrong...

Artificial intelligence is transforming the way we build applications, automate workflows, and...

Fable 5 was suspended for two weeks over a safeguard bypass, then redeployed with stronger guardrails and a proposed industry standard for scoring AI jailbreaks

New testimony reframes the Mythos export ban: a top general reportedly told a senator the model breached almost all classified systems in a red-team test, not i

A reverse-engineer found that Claude Code secretly changes tiny characters in the date it sends the model - a covert marker aimed at spotting resellers and copy

Here is a fact that should bother you more than it does: in a 2026 audit of 1,968 tasks drawn from...

AI might seem like a magic bullet for fixing security issues, but it's not that simple, warned Eugene...

Human-in-the-loop can improve AI safety, but it usually does not by default. Putting a person behind an approval button only helps when the consequence…

The attack doesn't require a compromise There's a new attack pattern getting documented...

Bank statement fraud in lending: borrowers edit balances, deposits and overdrafts before upload. Three structural signals — xref count, producer…

The story making the rounds is that Claude Code has been caught "steganographically marking" requests...

My last essay said every guardrail should be earned from a failure. For cheap, mechanical checks,...

10 Out of 11 Coding Agents Failed. Here's Why That Number Should Concern You. Researchers...

Discover how steganographic code injection exploits AI models like Claude to hide malicious payloads and what developers can do to defend against these sophisti

One time we opened a daemon's log on the box, saw that some secret values were mixed into it, and...

Ever wondered if the very thing that makes advanced AI models so smart could also be their Achilles'...

"What happened when I stopped designing my AI coach for flexibility and started designing it to be...

The article discusses the crucial issue of liability for AI-generated content, highlighted by a...

A novel Rust-based macOS implant, codenamed Gaslight, has been uncovered, distinguished by its unique...

Descubre y soluciona las vulnerabilidades más peligrosas que acechan a tus aplicaciones de IA para proteger tus datos y modelos.

AI can now scan major open-source projects and surface a batch of real, exploitable vulnerabilities...

Model distillation attacks let attackers replicate frontier AI capabilities without safety alignment. How logits-level filtering can defend against rogue distil

The EU AI Act enforcement starts August 2026. Article 13(3)(c) requires technical documentation of...

What happened after 2,000 people tried to hack my AI assistant Fernando Irarrázaval ran a challenge on hackmyclaw.com to see if anyone could leak secrets held b