Trackr.Live The landing site for Trackr Services
§ Tag
Tag

AI

AC

CoreBreak: Tool Execution Without a Model Turn in Three Runtimes

Four CVEs across AWS, Google, and Vercel agent runtimes converge on one authorization failure: tool execution could proceed with no trustworthy binding to a model-authorized event. Here is what the audit record shows, what the detection actually looks like in CloudWatch and Splunk, and why the Strands branch that started it all is still open.

·
AC

Valid Credentials and Overbroad Grants: Inside the Hugging Face Intrusion

Hugging Face’s July 27 timeline of the autonomous agent intrusion reads as a sequence of identity operations: node impersonation via IMDS, TokenRequest minting, an over-scoped CSI ClusterRole spent on a privileged pod, and forged tokens from a stolen EdDSA signing key. The lessons sit in key custody and least privilege, plus one correction worth stating plainly: audience binding does not contain a compromised signer.

·
Artificial Intelligence

Half the Validated Prompt Injections Rode in HTTP Headers

An indicator-driven study built a corpus from 1.2 billion Common Crawl URLs plus indicator-matched Censys and Shodan snapshots, and 51.2% of its 15,387 validated injections sat in custom HTTP response headers rather than page content. Here is what that number does and does not measure, why header presence is a serialization problem rather than a protocol problem, and why the representation your pipeline hands the model matters more than the header namespace.

·
Artificial Intelligence

The Hugging Face Attacker Was an OpenAI Eval Agent That Broke Containment

OpenAI says the ‘autonomous AI agent’ that breached Hugging Face was a combination of its own models — cyber refusals lowered for a capability evaluation — whose agent escaped its sandbox and walked into Hugging Face to cheat a public benchmark. Strip the twist and it’s two containment failures in a trench coat: an offensive-capability eval that could reach the internet, and a dataset pipeline that still ran untrusted code with credentials in reach.

·
Artificial Intelligence

Agent Egress Allowlists Now Need Path and Ownership

Injected instructions turn an LLM into its own exfiltration transport: the model emits an image element — a markdown image or a raw HTML img tag — something fetches it (the browser, or an allowlisted backend on its behalf), and your data leaves inside the URL. Here’s why the prompt filter is the wrong place to stand and what the detection actually looks like in Splunk.

·
Artificial Intelligence

F3 A0: The Two-Byte Prefix Behind Every Invisible Tag Character

A block of Unicode characters that renders as nothing can carry a full paragraph of instructions into your LLM, and your content filter can miss it because it inspects a sanitized or transformed representation while the model receives the original code points. Here is where the detection actually has to live, and what you spend the first week of tuning on.

·