/
Navigation
Chronicles
Browse all articles
Explore
Semantic exploration
Research
Entity momentum
Nexus
Correlations & relationships
Story Arc
Topic evolution
Drift Map
Semantic trajectory animation
Posts
Analysis & commentary
Pulse API
Tech news intelligence API
Browse
Entities
Companies, people, products, technologies
Domains
Browse by publication source
Handles
Browse by social media handle
Detection
Concept Search
Semantic similarity search
High Impact Stories
Top coverage by position
Sentiment Analysis
Positive/negative coverage
Anomaly Detection
Unusual coverage patterns
Analysis
Rivalry Report
Compare two entities head-to-head
Semantic Pivots
Narrative discontinuities
Crisis Response
Event recovery patterns
Connected
Search: /
Command: ⌘K
Embeddings: large
TEXXR

Chronicles

The story behind the story

days · browse · Enter similar · o open

To show an attack vector, a researcher made a package with a name hallucinated by ChatGPT that had 30K+ downloads in three months and appeared in many repos

Bar Lanyado / Lasso Security :

Lasso Security Bar Lanyado

Context & Ripple Effects

The demonstration extends an older package-registry weakness: attackers can exploit names that developers are likely to select by mistake, as shown by typo-package experiments that reached thousands of hosts. Here, the likely mistake originates in an AI-generated dependency suggestion rather than a keyboard typo.

The result also supplies an early concrete example of what later coverage termed “slopsquatting”: registering packages under plausible but nonexistent names produced by AI systems. Its download and repository presence show that a generated name can propagate beyond a single prompt.

First-order effects

  • Developers and repositories that accepted the hallucinated package name face an immediate dependency-selection risk: a malicious publisher could occupy the same kind of name before users verify whether it is a real library.
  • The demonstration gives security teams a testable signal to review—dependencies whose names originated in AI-assisted code or documentation suggestions, especially when provenance is unclear.

Second-order effects

  • Package registries, dependency scanners, and AI coding tools face pressure to make package identity and provenance easier to verify, rather than treating a syntactically plausible name as sufficient.
  • Attackers can shift from traditional typo targeting toward monitoring and registering likely AI-invented names, broadening the supply-chain attack surface without needing to compromise an established package.

Third-order effects

  • If AI-generated dependency recommendations become routine, open-source package trust will increasingly depend on provenance checks at installation and review time, not just on users recognizing familiar names.
  • This points to a wider open-source distribution trust problem: automation can speed legitimate development while also creating scalable paths for malicious packages to enter downstream projects.

The trend: AI-assisted development is creating a new software-supply-chain risk class in which model-generated identifiers can become attacker-controlled dependency names.

Discussion

  • @doctorow @doctorow on x
    These “hallucinations” are a stubbornly persistent feature of large language models, because these models only give the illusion of understanding. 4/
  • @doctorow @doctorow on x
    In reality, they are just sophisticated forms of autocomplete, drawing on huge databases to make shrewd (but reliably fallible) guesses about which word comes next: https://dl.acm.org/... 5/
  • @doctorow @doctorow on x
    Here's a fun AI story: a security researcher noticed that large companies' AI-authored source-code repeatedly referenced a nonexistent library (an AI “hallucination"), so he created a (defanged) malicious library with that name and uploaded it. 1/ [image]
  • @lassosecurity @lassosecurity on x
    🚨 Research alert Read @BLanyado follow-up research that dives deeper into AI Package Hallucination. Did #GPT4 #GEMINI #COHERE closed the security gaps? spoiler alert-no, is the attack effective in the wild? well- yes. For the full article➡️https:// www.lasso.security/...
  • @bruce_arthur Bruce Arthur on x
    The fundamental problem with AI, beyond the crushing soullessness and the computing power requirements, is that it's just not reliable. That's bad!
  • @blanyado @blanyado on x
    🚨 Research alert I just published my new research regarding LLM hallucinations. This time I asked A LOT more questions and investigated more models. I also found Hallucinated package in the wild with over 30K downloads of a hallucinated package https://www.lasso.security/...