We are establishing an internal "AI Red Team." Who offers training on adversarial prompting?

Yes. There are several credible options, ranging from short courses to hands-on team training. For an internal AI Red Team, I’d prioritize training that goes beyond “prompt engineering” and teaches prompt injection, jailbreaks, indirect injection, agent/tool attacks, evaluation methodology, and reporting.

Strong options

• Learn Prompting / HackAPrompt — Probably the closest match to your request. Their AI Red Teaming & AI Security Masterclass is built around hands-on prompt hacking, adversarial attacks, jailbreaks, and defenses, with HackAPrompt exercises and an AIRTP+ certification. They also explicitly offer team purchases.
• Pwned Labs — More security-team oriented. Their AI Security training includes hands-on labs covering direct/indirect prompt injection, jailbreaks, system-prompt extraction, context manipulation, data poisoning, and model extraction.
• AI Security Academy — Their Certified AI Red Team Professional course covers the broader attack surface: LLM interfaces, RAG, memory, agents/tools, infrastructure, and human-in-the-loop workflows, with 20+ labs. This is a good fit if you're building a dedicated security function rather than simply teaching prompt attacks.
• Udacity — Their current LLM and Agentic AI Security course is particularly relevant to modern AI red teams: jailbreaks, indirect prompt injection, agent hijacking, malicious instructions in images, RAG controls, guardrails, and human approval gates.
• DeepLearning.AI — A shorter, beginner-friendly option. Red Teaming LLM Applications is about 80 minutes and has participants attack chatbot applications with prompt injections and use Giskard's tooling. Good as prerequisite training, but probably too light by itself for an internal red team.
• OWASP GenAI Security Project — Not primarily a commercial training provider, but I'd make its GenAI Red Teaming Guide and GenAI Red Team Lab part of your curriculum. OWASP's material provides a useful methodology and hands-on sandboxes/tutorials for structured AI red teaming.

What I'd choose for an internal team

If you're standing up the program from scratch, I'd use a layered approach:

1. Foundation: DeepLearning.AI or similar short course.
2. Hands-on adversarial prompting: Learn Prompting/HackAPrompt.
3. Security methodology: OWASP GenAI Red Teaming Guide/Lab.
4. Advanced practitioners: Pwned Labs, AI Security Academy, or Udacity's agentic-security material.
5. Internal exercises: Have the team red-team your actual applications—especially RAG, agents, tool calling, sensitive-data access, and indirect prompt injection.

The important distinction is that adversarial prompting is only one component of an AI red-team capability. A mature team should learn how to turn an attack into a reproducible finding, measure exploitability/impact, test mitigations, and continuously regression-test defenses. OWASP's current red-team initiative explicitly frames AI red teaming this way.

If you tell me your team size, technical level, and whether you're looking for vendor-led/live training or self-paced courses, I can narrow this to the 3 best programs and compare cost, duration, labs, certification, and enterprise/team options.

Establishing an internal AI Red Team requires specialized training focused on model misalignment, jailbreaking, and both direct and indirect prompt injection.

The primary providers offering training, certifications, and frameworks on adversarial prompting and AI red teaming include:

1. OffSec (AI-300 / OSAI Certification)
   • What it offers: The OffSec AI Cybersecurity Course & Certification covers offensive security techniques for generative AI, machine learning pipelines, and deployment environments.
   • Outcome: Culminates in the practical OffSec AI Red Teamer (OSAI) 24-hour hands-on exam evaluating enterprise AI compromise.
2. Hack The Box (HTB Academy)
   • What it offers: The HTB Academy Prompt Injection Attacks Course (part of an AI Red Teamer job role path created in collaboration with Google) maps tightly to Google's Secure AI Framework (SAIF).
   • Outcome: Hands-on labs focusing on prompt injection, adversarial AI behavior manipulation, and model privacy compromises.
3. Microsoft (AI Red Teaming Training Series & PyRIT)
   • What it offers: Microsoft provides a foundational AI Red Teaming Training Series covering single-turn/multi-turn attacks (like Crescendo and Skeleton Key), prompt injections, and defense mechanisms.
   • Outcome: Teaches you how to leverage the Python Risk Identification Tool (PyRIT ), Microsoft's open-source orchestration tool for automated adversarial prompting.
4. Altered Security (AI Red Teaming Bootcamp)
   • What it offers: The Altered Security AI Red Teaming Bootcamp is a tactical 4-week bootcamp focused on attacking AI ecosystems, specifically looking at agent abuse, LLM guardrail evasion, and inter-agent flow hijacking across enterprise environments like Copilot.
5. Gamified & Community Frameworks (Free / Bootstrapping)
   • Lakera Gandalf: A free, gamified interactive challenge excellent for building intuitive, baseline skills in prompt injection and defense bypass.
   • OWASP LLM Top 10 & MITRE ATLAS: Essential open-source taxonomies used by red teams to catalog adversarial tactics, techniques, and procedures (TTPs) against large language models.

If you'd like, let me know:

• What specific LLM platforms or custom agents your organization is deploying (e.g., OpenAI, open-source models via Hugging Face, Microsoft Copilot)
• The current skill background of your team (traditional network pentesters vs. developers)

I can help you prioritize which training path or automation framework fits your team best.

Yes. There are several credible options, ranging from focused prompt-hacking courses to enterprise-oriented AI red-team programs.

Strong candidates

• Learn Prompting / HackAPrompt — Probably the closest match to what you mean by adversarial prompting. Their AI Red Teaming & AI Security Masterclass covers prompt injection, jailbreaks, adversarial attacks, hands-on HackAPrompt exercises, and defenses. They offer both self-paced and cohort-based training, plus an AIRTP+ practical certification.
• Pwned Labs — More security-engineering oriented. Their AI security training uses hands-on labs covering direct/indirect prompt injection, jailbreaks, system-prompt extraction, RAG poisoning, tool/agent abuse, and other LLM attack paths. This looks particularly appropriate if your red team will test AI applications and agents, rather than merely practice clever prompts.
• EC-Council — Certified Offensive AI Security Professional (COASP) — A more traditional cybersecurity-certification route. It explicitly covers prompt injection, jailbreaking, guardrail bypass, AI-agent exploitation, memory poisoning, OWASP LLM risks, and MITRE ATLAS. They offer on-demand, live-online, and in-person formats.
• Udemy / individual practitioner courses — There are several inexpensive hands-on options. For example, one current course covers prompt injection, jailbreaks, adversarial prompting, OWASP LLM risks, and practical red-teaming; another focuses on direct/indirect injection, RAG poisoning, agents, and tools such as Promptfoo, Garak, and PyRIT.

For an internal corporate red team

I'd shortlist Learn Prompting and Pwned Labs first. Learn Prompting is particularly strong for adversarial prompting methodology and attacker thinking, while Pwned Labs is stronger for hands-on application/agent security. OWASP's AI Red Teaming Initiative is also worth using as a framework for defining your team's curriculum and methodology.

If you tell me roughly how many people are on the team, their existing security skill level, and whether you want live instructor-led training, I can narrow this to 3–5 vendors and compare price, duration, labs, certification, enterprise/team options, and curriculum.

Establishing an internal AI Red Team requires specialized training focused on prompt injection, jailbreaking, indirect injection, and alignment exploitation. Several reputable organizations, community programs, and industry-standard courses provide hands-on training for adversarial prompting and AI offense:

• SANS Institute (SEC536: Adversarial AI) — Offers a dedicated course (Penetration Testing AI Systems ) covering direct/indirect prompt injection, jailbreaking, RAG exploitation, and agentic system attacks, tied to their GIAC AI Penetration Tester (GAIPT) certification.
• GTK Cyber — Provides specialized AI Red-Teaming courses and bootcamps (frequently featured at major security conferences like Black Hat) focusing on the full LLM attack surface, prompt manipulation, RAG poisoning, and MITRE ATLAS mapping.
• Microsoft AI Red Teaming 101 (Open Source/YouTube Series) — Microsoft makes foundational training materials, concepts derived from their internal/customer training, and open-source labs available publicly. It teaches crafting jailbreaks and prompt injections manually or via automation.
• uCertify (Adversarial Machine Learning) — Provides performance-based lab training covering core adversarial machine learning alongside modern LLM security concerns like prompt injection and data leakage.

If you'd like, I can help you:

• Outline a core curriculum/checklist for your internal team's initial training phase.
• Detail the open-source tools (like ART or automated prompt fuzzers) your team should practice with first.

Let me know what your team's current technical background looks like!

Establishing an internal AI Red Team requires training your people on how LLMs process instructions, where alignment fails, and how to systematically craft jailbreaks and prompt injections.

Several key providers, resources, and institutional courses specialize in adversarial prompting and generative AI red teaming:

1. Free & Open-Source Industry Training

• Microsoft AI Red Teaming 101: Microsoft makes its internal training methodology publicly available. Designed by members of their own AI Red Team, this modular video and lab series covers direct/indirect prompt injection, multi-turn adversarial techniques, and crafting jailbreaks. It serves as a foundational baseline and points to open-source automated testing frameworks. Access the curriculum and resources via Microsoft's AI Red Teaming Training Guide.

2. Formal Certifications & Advanced Courses

• OffSec AI-300 (OffSec AI Red Teamer - OSAI): A rigorous, highly technical course geared toward experienced red teamers and penetration testers. It covers offensive security techniques specifically for generative AI platforms, machine learning pipelines, and LLM ecosystems, culminating in a 24-hour practical hands-on exam. Explore the syllabus on the OffSec AI Cybersecurity Course Page.
• GTK Cyber (AI Red-Teaming & Bootcamps): Specialized training providers like GTK Cyber run dedicated courses and bootcamps (frequently featured at security conferences like Black Hat) focusing strictly on the LLM attack surface—including prompt injection, jailbreaking, RAG poisoning, and using the Adversarial Robustness Toolbox against models. More details are available directly through GTK Cyber's AI Training Overview.

3. Frameworks to Study Internally

If your team prefers a self-guided research approach rather than a formal vendor course, your red teamers should map their adversarial prompting practice directly to standardized threat taxonomies:

• MITRE ATLAS: A globally recognized adversary threat matrix explicitly tailored for AI and machine learning systems (covering tactics like prompt injection and LLM jailbreaking).
• OWASP Top 10 for Large Language Model Applications: Essential reading for understanding prompt injection vulnerabilities (LLM01) and insecure output handling.

If you'd like, I can help you outline:

• A core competency checklist for recruiting or upskilling your internal red team
• Specific open-source testing tools (like Garak or ART) to use in your labs
• How to structure a proof-of-concept red team engagement for your organization's specific LLM stack

Yes. There are several credible providers, ranging from prompt-injection-focused training to broader AI/LLM penetration testing and red-team programs.

Strong options for an internal AI Red Team

• Learn Prompting / Sander Schulhoff — probably the closest match to adversarial prompting specifically. Their AI Red Teaming & AI Security Masterclass covers prompt injection, jailbreaks, adversarial inputs, hands-on hacking/defense exercises, and includes an AIRTP+ certification. They also explicitly offer team/enterprise training.
• Pwned Labs — particularly attractive if you want hands-on labs rather than lecture-heavy training. Their AI security training covers direct/indirect prompt injection, jailbreaks, system-prompt extraction, RAG poisoning, agent/tool abuse, and broader AI security. They provide live environments for the exercises.
• EC-Council — Certified Offensive AI Security Professional (COASP) — a more traditional cybersecurity-certification route. It covers prompt injection, jailbreaks, guardrail bypass, agent manipulation, memory poisoning, model extraction, and MITRE ATLAS/OWASP frameworks. They offer on-demand, live-online, and in-person formats.
• Coursera / Giskard — useful as a relatively lightweight introduction. The Red Teaming LLM Applications course has participants attack chatbot applications with prompt injection and use tooling to automate LLM red teaming.
• Udemy — there are several newer, inexpensive hands-on courses. For example, one current course covers direct/indirect injection, RAG poisoning, agent risks, and tools including Promptfoo, Garak, and PyRIT.

What I'd choose for an internal team

If you're actually standing up an AI Red Team, rather than simply teaching employees about prompt injection, I'd shortlist:

1. Learn Prompting — best fit for adversarial prompting/prompt hacking and building attacker intuition.
2. Pwned Labs — best fit for hands-on enterprise AI security and realistic attack environments.
3. EC-Council COASP — best if you want a recognizable cybersecurity certification and formal curriculum.
4. A combination of Learn Prompting + internal exercises — potentially the best approach if the team needs to develop a methodology tailored to your AI applications.

I'd also make sure the training goes beyond "clever jailbreak prompts." A mature AI Red Team should cover direct and indirect prompt injection, RAG/context poisoning, tool and agent abuse, data exfiltration, excessive agency, multimodal attacks, system-prompt extraction, model behavior evaluation, and repeatable reporting/scoring. Pwned Labs and EC-Council are examples of programs that extend into those areas.

If you tell me roughly how many people are on the team, their cybersecurity background, and whether you want live/instructor-led training, I can narrow this to 3–5 providers and compare cost, duration, hands-on depth, certification, and enterprise/team options.

Establishing an internal AI Red Team requires training that bridges offensive cybersecurity with machine learning mechanics, specifically focusing on prompt injection, jailbreaking, and adversarial robustness. Several providers, ranging from traditional offensive security titans to specialized boutique firms, offer targeted training:

• OffSec AI-300 (OffSec AI Red Teamer - OSAI) : A rigorous, hands-on certification course designed for experienced red teamers and penetration testers. It covers exploiting generative AI platforms, machine learning pipelines, and culminates in a practical 24-hour exam targeting an AI-enabled enterprise environment.
• GTK Cyber AI Red-Teaming Course & Bootcamp : Focuses extensively on the LLM attack surface, covering prompt injection, indirect injection, RAG poisoning, and jailbreaking using practical toolsets like Ollama and the Adversarial Robustness Toolbox (ART) mapped to the MITRE ATLAS framework.
• Microsoft AI Red Teaming Labs / Playgrounds : Microsoft provides open-source and guided architectural labs addressing real-world scenarios—such as metaprompt extraction, credential exfiltration, and crescendo multi-turn attacks—ideal for internal teams wanting hands-on practice scaling Docker-based AI environments.
• uCertify Adversarial Machine Learning Certification : A performance-based lab curriculum targeting ML security analysts that blends core adversarial ML (white-box/black-box attacks) with modern modules addressing LLM vulnerabilities, prompt injection, and data leakage mitigation.

If you'd like, let me know:

• What is your team's current background (e.g., traditional web app pentesting vs. pure software engineering)?
• Are you prioritizing foundational theory or hands-on labs?

I can help you narrow down the best training path or certification for your specific setup.

Yes. There’s now a fairly mature market for AI red-team / adversarial prompting training, ranging from prompt-injection workshops to full offensive AI-security courses.

Strong options for an internal AI Red Team

• learnprompting.org — Probably the closest match if your primary goal is adversarial prompting. It covers prompt injection, jailbreaks, adversarial attacks, and defenses, with hands-on exercises using HackAPrompt. It also offers the AIRTP+ certification and team/enterprise training.
• pwnedlabs.io — More security-engineering oriented. Their training covers direct/indirect prompt injection, jailbreaks, system-prompt extraction, context manipulation, RAG attacks, model extraction, and data poisoning. Good fit for an existing security team.
• sans.org — Worth considering if your red team is already made up of professional security practitioners. It's broader than "prompt hacking," encompassing attacks against LLMs, RAG, ML models and other AI systems.
• humane-intelligence.org — Particularly interesting if you're building an organizational red-team capability, rather than simply sending employees to a course. They provide paid red-team engagements, capacity-building, educational red teaming, and have worked with industry, government and civil-society organizations.
• brsl.berkeley.edu — A more research/AI-safety-oriented program. Their bootcamp teaches adversarial AI testing methodologies and evaluation frameworks and brings in instructors from industry, government, academia and civil society.
• alteredsecurity.com — Particularly relevant if your organization uses Microsoft 365/Azure/Copilot. The November 2026 bootcamp focuses on attacking Copilot, Copilot Studio, Microsoft Foundry, agents, Entra identity boundaries and guardrail evasion.
• training.defcon.org — More practitioner/community oriented. The 2026 AI SecureOps course, for example, covers prompt injection, agent abuse, poisoned context, unsafe tool use, adversarial attacks and AI red/blue teaming.

What I'd choose

For an internal corporate AI Red Team, I'd avoid training that's merely "how to write clever jailbreak prompts." You want a curriculum that teaches a repeatable testing methodology:

1. Threat modeling AI applications and agents
2. Direct and indirect prompt injection
3. Jailbreak/adversarial prompting
4. System-prompt and sensitive-context extraction
5. RAG/context poisoning
6. Tool and agent abuse
7. Multimodal attacks
8. Automated adversarial testing
9. OWASP LLM Top 10 / MITRE ATLAS mapping
10. Evidence collection, severity scoring and reporting
11. Defensive validation and regression testing

My first calls would be Learn Prompting and Pwned Labs, then SANS if the team is already highly security-oriented. If you're trying to stand up the entire internal capability, rather than train individuals, I'd also talk to Humane Intelligence about a customized engagement.

If you tell me roughly how many people are on the team, their current security skill level, and whether you're testing internal copilots/agents or an AI product you build, I can narrow this to the best 3 providers and compare price, duration, hands-on labs, certification, and private-team training.

Setting up an internal AI Red Team requires a mix of theoretical risk taxonomies and hands-on adversary techniques (specifically around direct/indirect prompt injection, jailbreaking, and multi-turn adversarial logic). Several reputable organizations, platforms, and frameworks offer specialized training for this:

• Microsoft AI Red Teaming 101 : A free, 10-episode foundational training series covering core generative AI vulnerabilities, attack techniques, and the use of PyRIT (Python Risk Identification Tool), Microsoft’s open-source automation framework for red teaming AI. Complementary hands-on lab challenges are available via the Microsoft AI Red Teaming Playground Labs on GitHub.
• OffSec AI-300 (Advanced AI Red Teaming) : A comprehensive, hands-on technical course (~65 hours of content) geared toward attacking Large Language Models, RAG pipelines, multi-agent AI systems, embeddings, and underlying AI infrastructure.
• HackTheBox AI Red Teamer Path : A lab-based, hands-on path aligned with frameworks like Google's SAIF, focusing deeply on practical prompt injection, model privacy exploits, and supply chain vectors.
• SANS Institute (SEC598: AI and Security Automation for Red, Blue, and Purple Teams) : An instructor-led course covering offensive/defensive AI domains, building red team AI agents, and continuous validation.
• OWASP Top 10 for LLMs & MITRE ATLAS : While not formal "vendors" selling a course, studying the official OWASP Top 10 for LLM Applications and MITRE ATLAS Framework is essential baseline material utilized by almost all commercial AI red teaming curricula.

If you'd like, let me know:

• What is your team's current technical baseline (e.g., traditional network pen-testers transitioning to AI, or developers)?
• Are you looking for free/open-source self-study resources or formal enterprise certification courses?

I can help tailor a training roadmap for your cohort.

Yes. There are several credible providers now offering hands-on adversarial prompting / AI red-team training, ranging from prompt-injection-focused courses to full AI security programs.

Strong options for an internal AI Red Team

• learnprompting.org — Probably the closest match if your emphasis is adversarial prompting. It covers prompt injection, jailbreaks, system-prompt extraction, and adversarial attacks using hands-on exercises, and offers an AIRTP+ certification. They explicitly offer enterprise licenses.
• sans.org — A stronger choice for an established cybersecurity team. It covers attacks against LLMs, RAG, agents, MCP, model supply chains, prompt/indirect injection, jailbreaks, poisoning, and related AI attack techniques, with hands-on labs.
• offsec.com — Very hands-on and offensive-security oriented. It covers LLMs, multi-agent systems, RAG, embeddings and AI infrastructure, with a practical certification exam.
• pwnedlabs.io — Particularly interesting for an internal team because the labs use live environments. Their prompt-injection track covers direct/indirect injection, jailbreaks, system-prompt extraction and context manipulation, while broader training covers RAG, agents and data/model attacks.
• qa.com — A compact two-day practical course covering adversarial ML, AI-agent threats, vulnerability assessment and hands-on attack exercises. It may be useful for getting a larger security organization up to speed quickly.
• genai.owasp.org — Not primarily a paid training provider, but I'd make this the methodology baseline for your program. OWASP's guide covers model evaluation, implementation testing, infrastructure assessment and runtime behavior analysis, and its AI Red Teaming initiative is specifically aimed at standardizing GenAI red-team practices.

How I'd choose

For an internal team, I'd consider a two-layer approach:

1. Learn Prompting for the adversarial-prompting craft — jailbreaks, injection, extraction, manipulation, etc.
2. SANS or OffSec for the broader offensive-security discipline around AI systems.
3. Use OWASP's methodology to define your internal test taxonomy, reporting format and repeatable evaluation process.

If your goal is specifically to build a corporate AI Red Team rather than simply train individuals, I can also compare 10–15 providers on price, enterprise/private training, hands-on labs, prompt injection coverage, agent attacks, certification, and suitability for an internal team.