Search

Find a vulnerability

Search criteria ⓘ Use this form to refine search results.
Full-text search supports keyword queries with ranking and filtering.
You can combine vendor, product, and sources to narrow results.
Enable “Apply ordering” to sort by date instead of relevance.

    1 vulnerability by OpenAI GPT-2

    AVID-2023-V008

    Vulnerability from avid – Published: 2023-03-31 – Updated: 2023-03-31 ATLAS Case Study
    Summary
    OpenAI built GPT-2, a language model capable of generating high quality text samples. Over concerns that GPT-2 could be used for malicious purposes such as impersonating others, or generating misleading news articles, fake social media content, or spam, OpenAI adopted a tiered release schedule. They initially released a smaller, less powerful version of GPT-2 along with a technical description of the approach, but held back the full trained model. Before the full model was released by OpenAI, researchers at Brown University successfully replicated the model using information released by OpenAI and open source ML artifacts. This demonstrates that a bad actor with sufficient technical skill and compute resources could have replicated GPT-2 and used it for harmful goals before the AI Security community is prepared.
    Risk domain
    Security
    SEP view
    S0502: Model theft
    Lifecycle
    L04: Model Development, L06: Deployment
    Organisations
    OpenAI GPT-2 (deployer)
    Affected artifacts
    Artifact Type
    OpenAI GPT-2 System
    References
    URL Label
    https://atlas.mitre.org/studies/AML.CS0007 GPT-2 Model Replication
    https://www.wired.com/story/dangerous-ai-open-source/ Wired Article, "OpenAI Said Its Code Was Risky. Two Grads Re-Created It Anyway"
    https://blog.usejournal.com/opengpt-2-we-replicat… Medium BlogPost, "OpenGPT-2: We Replicated GPT-2 Because You Can Too"
    To clipboard 

    {
  "affects": {
    "artifacts": [
      {
        "name": "OpenAI GPT-2",
        "type": "System"
      }
    ],
    "deployer": [
      "OpenAI GPT-2"
    ],
    "developer": []
  },
  "credit": null,
  "data_type": "AVID",
  "data_version": "0.2",
  "description": {
    "lang": "eng",
    "value": "OpenAI built GPT-2, a language model capable of generating high quality text samples. Over concerns that GPT-2 could be used for malicious purposes such as impersonating others, or generating misleading news articles, fake social media content, or spam, OpenAI adopted a tiered release schedule. They initially released a smaller, less powerful version of GPT-2 along with a technical description of the approach, but held back the full trained model.\n\nBefore the full model was released by OpenAI, researchers at Brown University successfully replicated the model using information released by OpenAI and open source ML artifacts. This demonstrates that a bad actor with sufficient technical skill and compute resources could have replicated GPT-2 and used it for harmful goals before the AI Security community is prepared.\n"
  },
  "impact": {
    "avid": {
      "lifecycle_view": [
        "L04: Model Development",
        "L06: Deployment"
      ],
      "risk_domain": [
        "Security"
      ],
      "sep_view": [
        "S0502: Model theft"
      ],
      "taxonomy_version": "0.2"
    }
  },
  "last_modified_date": "2023-03-31",
  "metadata": {
    "vuln_id": "AVID-2023-V008"
  },
  "problemtype": {
    "classof": "ATLAS Case Study",
    "description": {
      "lang": "eng",
      "value": "GPT-2 Model Replication"
    },
    "type": "Advisory"
  },
  "published_date": "2023-03-31",
  "references": [
    {
      "label": "GPT-2 Model Replication",
      "type": "source",
      "url": "https://atlas.mitre.org/studies/AML.CS0007"
    },
    {
      "label": "Wired Article, \"OpenAI Said Its Code Was Risky. Two Grads Re-Created It Anyway\"",
      "type": "source",
      "url": "https://www.wired.com/story/dangerous-ai-open-source/"
    },
    {
      "label": "Medium BlogPost, \"OpenGPT-2: We Replicated GPT-2 Because You Can Too\"",
      "type": "source",
      "url": "https://blog.usejournal.com/opengpt-2-we-replicated-gpt-2-because-you-can-too-45e34e6d36dc"
    }
  ],
  "reports": null
}