feat: add Hallucination vulnerability with fake_citations, fake_apis,…

[sayan5069]

Summary

Adds a Hallucination vulnerability class to pair with the existing
HallucinationMetric in metrics/.

Problem

HallucinationMetric existed in metrics/ but had no corresponding
vulnerability in vulnerabilities/ to generate adversarial attacks for it.

Changes

• Added deepteam/vulnerabilities/hallucination/types.py
• Added deepteam/vulnerabilities/hallucination/template.py
• Added deepteam/vulnerabilities/hallucination/hallucination.py
• Added deepteam/vulnerabilities/hallucination/__init__.py
• Registered Hallucination in deepteam/vulnerabilities/__init__.py

Vulnerability Types

• fake_citations - fabricated academic papers/sources
• fake_apis - fabricated SDK methods/endpoints
• fake_entities - fabricated people/companies/products
• fake_statistics - fabricated numerical data/surveys

Standards Mapping

• OWASP LLM09 (Misinformation)

Usage

from deepteam.vulnerabilities import Hallucination
hallucination = Hallucination(types=["fake_citations", "fake_apis"])
risk_assessment = red_team(model_callback=callback, vulnerabilities=[hallucination], attacks=[PromptInjection()])

[vercel]

@sayan5069 is attempting to deploy a commit to the Confident AI Team on Vercel.

A member of the Team first needs to authorize it.

[penguine-ip]

Hey @sayan5069 thanks for the PR! Do you mind adding the docs as well? @A-Vamshi on this one

[sayan5069]

Hi @penguine-ip, just added the docs! Let me know if anything needs changing.

[A-Vamshi]

Can we remove this section and replace it with a mermaid diagram as we have for the other docs? Here's an example: https://github.com/confident-ai/deepteam/blob/main/docs/docs/red-teaming-vulnerabilities-bias.mdx?plain=1#L80

You can just copy paste the code block for mermaid diagram and change the names for vulnerability, that would be enough, thanks!

[A-Vamshi]

Can you format the lines for the tables here? (Space them correctly to look like a table in the code too, here's an example: https://github.com/confident-ai/deepteam/blob/main/docs/docs/red-teaming-vulnerabilities-bias.mdx?plain=1#L64). Not a major issue but would be cleaner and more easier to maintain going forward

[A-Vamshi]

Hey @sayan5069, this PR overall looks solid, great work! Just a few more minor things:

• Can you add this vulnerability to the constants of the vulnerabilities? Here's the file I'm talking about: https://github.com/confident-ai/deepteam/blob/main/deepteam/vulnerabilities/constants.py
• Can you add some tests for this vulnerability as well? Here's an example you can use as reference: https://github.com/confident-ai/deepteam/blob/main/tests/test_core/test_vulnerabilities/test_bias.py
• And finally just added some comments in the PR itself, could you please look into them?
  Just these minor changes and this PR will be good to merge, once again, thanks for the PR, awesome work!

[sayan5069]

Hi @A-Vamshi, done! Added Hallucination to constants.py, added tests, and fixed the docs — updated table formatting and replaced Standards Mapping with a mermaid diagram. Let me know if anything else needs changing!

[sayan5069]

Hi @A-Vamshi @penguine-ip, just checking in — let me know
if there's anything else needed from my side!