Safety Is Becoming a Key Factor for AI
Leading AI providers such as OpenAI and Anthropic are making one thing increasingly clear: trust in artificial intelligence cannot be built on performance alone. As a result, safety, transparency and traceability are moving further into focus for companies and organisations.
The discussion around artificial intelligence is increasingly shifting from pure performance towards trust, safety and independent verification. Recent statements from OpenAI and Anthropic demonstrate that AI governance, safety and independent evaluation are becoming essential prerequisites for the responsible use of AI.
While OpenAI highlights the challenges surrounding AI safety and alignment as a strategic success factor, Anthropic is calling for stronger external evaluation of powerful AI systems through the concept of independent “Embedded Evaluators”. The message is clear: trust in AI should not rely solely on the claims of its developers.
“The more responsibility we transfer to AI, the less sufficient it becomes for developers to assess their own systems. Trust requires independent evidence,” says Michael Weberberger, Managing Director of TRUSTIFAI.
From Governance to Independent Assurance
As AI becomes increasingly embedded in business-critical processes, the requirements for transparency, traceability and risk management are growing. Companies must not only define how AI may be used, but also be able to demonstrate that systems actually meet the requirements that have been established.
AI governance provides the organisational framework, AI safety implements safeguards, and testing verifies system behaviour. In addition, AI assurance is becoming increasingly important: the independent evaluation of risks, controls and system behaviour.
“Good AI governance defines the right rules. What ultimately matters, however, is whether we can also demonstrate that AI systems actually comply with these rules in practice,” says Weberberger.
How TRUSTIFAI Supports Companies
As an independent specialist for trustworthy AI, TRUSTIFAI – a joint venture between TÜV AUSTRIA and the Software Competence Center Hagenberg (SCCH) – supports companies throughout the entire lifecycle of AI systems. Its portfolio of services includes:
- Independent AI assessments and audits
- Evaluation of AI risks and safety measures
- Validation of governance and compliance requirements
- Robustness analyses
- Preparation for regulatory requirements such as the EU AI Act
- Building robust evidence for internal and external stakeholders
The objective is not merely to claim that AI can be trusted, but to make that trust demonstrable.
“Independent evaluation is increasingly evolving from a regulatory requirement into a business necessity. Companies need robust evidence of how their AI systems work, what risks exist and how effective their safeguards are. This is exactly where TRUSTIFAI comes in,” explains Weberberger.
Developments at OpenAI and Anthropic indicate that the industry is moving towards a new understanding of trustworthy AI. As AI systems become more powerful and increasingly autonomous, independent verification of safety, robustness and compliance is gaining further importance.
“Companies that want to scale AI successfully need more than powerful technology. They also need robust evidence that their AI can be trusted,” concludes Michael Weberberger.