LIVE From TECHEX - AI & Big Data Expo Global

AI Investment and Market Stage [00:00]

• 18% of all Fortune 500 companies are currently deploying budgets on AI implementation.
• Over 50% of current global revenue in AI is derived from services, signaling that productization in AI is still very early.

"18% of all Fortune 500 companies budgets it budgets are now deployed on AI... 18% so doesn't it make sense that some of it should be put for uh tools like yours"

Regulatory Landscape and AI Governance [00:43]

• As AI moves towards productization, there's a need for regulation and a "Golden Rule," which is currently missing.
• The current regulatory environment is fragmented, with states having their own rules, alongside existing compliance frameworks like HIPAA, CCPA, GDPR, the Biden Act, and the European New Y Act, leading to a "Wild West" scenario.
• The recommendation is to regulate AI in a way that enables public and companies to use it correctly, rather than making it forbidden, emphasizing testing, shadow AI governance, and AI security.

"for me it's the Wild Wild West saying again and I think we should regulate it in a sort of way but we should not for make it forbidden we should enable actually the range public and the companies to use it the right way"

Competitive Landscape and Third-Party Tools [01:42]

• Concerns exist about large language models (LLMs) or major players developing competitive tools, potentially squeezing smaller providers.
• A good CISO should always consider third-party tools to protect sensitive applications, including AI, to gain an objective opinion.
• Relying 100% on a single provider for infrastructure and security, including AI, doesn't make sense; an unbiased, objective opinion from a third party is crucial.
• The current LLM space is not a "winner takes it all" market, suggesting distributed models and the use of various LLMs (ChatGPT, Llama, open-source) will persist.
• Third-party tool sets and partners can help mitigate risks associated with AI.

"why would you buy infrastructure and security 100% from the same provider it doesn't make sense to me right I want to have an unbiased objective opinion and that's exactly what happened in the cloud security space"

Agentic AI and Platform Customization [03:22]

• Agentic AI is a significant emerging trend, and it can harbor similar vulnerabilities to existing AI systems.
• The platform is designed for customization, allowing organizations (e.g., banks) to define their own test cases, red teaming scenarios, and boundaries for AI interactions, including conversation restrictions and data poisoning concerns.
• The future involves rapid change monitoring, assessing guard rails, aggression limits, and staying ahead of new attacks, especially concerning AI agents.

"we are very much thinking about agentic security because this is the future as you say but it's very early in the days"

The Importance of AI Red Teaming and IP [05:27]

• The core of the business is AI red teaming and research, which serves as a unique selling proposition (USP), with experience in this area for two years.
• The company possesses its own intellectual property within the organization, with the Splick platform's IP fully owned and based in the United States.

"the core of the business is the AR red team at the research right yeah that's the USB even if you ask me the Head Start been doing this since two years and the USP with the red team actually"

Actionable Advice for AI Practitioners [06:05]

• Practitioners in the AI space should not overly worry about compliance and governance in the initial stages, as aspects like shadow AI and internal AI applications are already present.
• Implementing AI firewalls requires careful testing to avoid over-aggression, which could disclose or neutralize key functionalities of chatbots or applications.
• The hardest part of the business is testing domain-specific applications, not the models themselves, as it requires speaking the industry's language and adapting to its rapid evolution.

"the testing piece of the main specific applications not models itself because the domain specific means you need to talk in the language of the chat but and the industry in the vertical is the hardest part of this business because it's evolving so fast"