← Back to Overview

Data & Analysis →

---

Safety and Alignment: Sources

Model Safety Documentation

Anthropic

1. Anthropic Transparency Hub: Model Report
   • URL: https://www.anthropic.com/transparency/model-report
   • Key Data: Safety evaluations, prompt injection resistance
2. Claude Opus 4.5 Safety Features
   • URL: https://www.anthropic.com/news/claude-opus-4-5
   • Published: November 2025
   • Key Data: Industry-leading prompt injection resistance
3. Constitutional AI Paper
   • Type: Academic research
   • Key Data: Anthropic’s safety training methodology

OpenAI

1. GPT-5 System Card
   • Type: Technical documentation
   • Key Data: Safety evaluations and mitigations
2. OpenAI Red Teaming Network
   • URL: https://openai.com/index/red-teaming-network/
   • Key Data: Adversarial testing methodology

Google DeepMind

1. Gemini Safety Documentation
   • URL: https://deepmind.google/models/gemini/
   • Key Data: Safety features and testing
2. Gemini 3 Safety Report
   • Type: Technical documentation
   • Key Data: Multi-modal safety evaluations

---

Safety Research

Prompt Injection

1. Prompt Injection Attacks and Defenses
   • Type: Academic literature
   • Key Topics: Attack vectors, mitigation strategies
2. LLM Security Research
   • Type: Academic literature
   • Key Topics: Adversarial attacks on language models

Hallucination

1. TruthfulQA Benchmark
   • Type: Academic benchmark
   • Key Data: Factual accuracy evaluation methodology
2. Hallucination in LLMs: Survey
   • Type: Academic survey
   • Key Topics: Types, causes, mitigation

Jailbreaking

1. Jailbroken: How Does LLM Safety Training Fail?
   • Type: Academic research
   • Key Data: Jailbreak techniques and success rates
2. HarmBench: A Standardized Evaluation Framework
   • Type: Academic benchmark
   • Key Data: Harmful output evaluation

---

Industry Coverage

Safety Analysis

1. VentureBeat: Claude Safety Analysis
   • URL: https://venturebeat.com/ai/anthropics-claude-opus-4-5-is-here-cheaper-ai-infinite-chats-and-coding
   • Published: November 2025
   • Key Data: Safety comparison
2. Data Studios: Claude 4 Safety Benchmarks
   • URL: https://www.datastudios.org/post/claude-4-in-2025-performance-safety-benchmarks-ecosystem-news-and-real-world-impact-for-enterpr
   • Key Data: Safety benchmark analysis
3. The Great AI Privacy Divide
   • URL: https://medium.com/@michael_79773/the-great-ai-privacy-divide-one-year-later-two-worlds-apart-e74ce6187f1f
   • Key Topics: Privacy and safety considerations

---

Policy and Governance

AI Safety Policy

1. UN AI Safety Discussions
   • Type: Policy documents
   • Key Topics: International AI governance
2. OECD AI Principles
   • Type: Policy framework
   • Key Topics: Responsible AI development

Academic Programs

1. AI Safety Research Programs
   • Type: Academic landscape
   • Key Topics: University research initiatives
2. AI Ethics Curriculum Resources
   • Type: Educational materials
   • Key Topics: Teaching AI safety

---

Source Categories

Category	Count
Model documentation	7
Safety research	6
Industry coverage	3
Policy/governance	4
Total	20