FOR business leaders, the relentless drive for profit, speed, and efficiency has traditionally framed artificial intelligence as an enabler. Consequently, the AI conversation often centres on capability: faster models, larger datasets, and higher productivity.

Today, however, the critical question is responsibility. Responsible AI begins with human dignity, not technical performance. It demands transparency about limitations, independent auditing, refusal of dangerous dual-use requests, and human accountability for high-stakes decisions. Conversely, harmful AI hides intent, evades safeguards, and enables weapons, surveillance, or disinformation.

As market conditions evolve, driving innovation becomes paramount. Yet AI responsibility must serve as innovation’s constant companion. Sustainable leadership empowers organisations to address real market needs while operating responsibly. To build an enduring enterprise, leaders must ensure their creations earn the respect of the global market. It is the board's duty to align the entire organisation around both performance targets and ethical obligations.

Below are practical observations and case studies that turn this conversation from an abstract manifesto into an executive guide.

Consider the human cost of unchecked AI:

Wrongful arrests: A Tennessee grandmother spent six months in jail after police relied on a faulty facial recognition match to charge her with bank fraud. In Florida, a commercial crabber was arrested for a crime in a city he had never visited because an algorithm returned a 93% match based on poor-quality surveillance footage.

Algorithmic bias: Workday, a major HR software provider, faces a class-action lawsuit alleging its AI screening tools systematically discriminated against Black, disabled, and older job applicants.

These are not peripheral cases. They represent real people whose lives were upended by algorithms that lacked proper oversight.

In May 2026, Google confirmed that one of its Gemini models inadvertently breached the systems of three real companies during a cybersecurity evaluation. Assigned to a simulated target that shared a name with an actual company — and with internet access left unintentionally open — the model guessed passwords and accessed live corporate services using credentials found in public repositories.

Although Google reported no damage and compared the incident to a "bug bounty" program, security experts raised immediate alarms. As Jack Cable, CEO of AI security startup Corridor, noted: "Models are going outside the bounds of what they should be doing, and conducting actual cyberattacks."

This episode raises a sobering question for every board: If your AI vendor’s models can accidentally breach real companies during a test, what happens when they are deployed without guardrails?

The threat extends directly to financial markets. Shareholders recently sued Ostin Technology Group over an alleged US$950 million pump-and-dump scheme that used AI-generated deepfakes of prominent leaders — including Goldman Sachs’ chief strategist, Elon Musk, and Mark Zuckerberg — to lure investors via sponsored social media ads. The stock surged 1,175% before crashing 94% in a single day.

Harmful AI hides intent, evades controls, and enables market-moving disinformation at scale.

Responsibility cannot rest solely on corporations; governments must establish clear boundaries. The EU AI Act imposes significant obligations on high-risk and general-purpose models, while the US Executive Order seeks to centralise oversight by preempting state laws. Both converge on a single reality: fragmented governance is incompatible with credible AI oversight.

Researchers need oversight, the public needs transparency and education, and end-users must be empowered to ask critical questions:

Who benefits?

Who is held accountable?

Are safeguards independently verifiable?

Responsible AI answers these questions openly. Harmful AI avoids them.

Leaders must recognise that users cannot judge an AI system's safety or intent from its outputs alone. Trust must rest on institutional frameworks rather than personal detection.

"Good AI" demonstrates transparency through published safety reports, independent audits, robust safeguards, and verifiable provenance. For instance:

Red teaming and evaluation: Credit Karma utilises red teams to test LLM workflows for vulnerabilities, running them through a five-step evaluation framework covering accuracy, safety, compliance, and data provenance.

Model documentation: Vendors should be required to provide standardised "model cards" documenting intended use cases, known limitations, and training data sources prior to deployment.

Proactive restraint: Anthropic’s 2026 report highlighted that its Claude models were cited in research that could potentially aid biological weapons development. Because legitimate dual-use vaccine research and weaponisation can appear identical, Anthropic blocked the accounts and tightened safety controls. That level of caution exemplifies responsible AI in practice: when consequences are severe, uncertainty must never be treated as permission.

AI risk is not merely an ethical debate—it is an operational, legal, reputational, and geopolitical issue. Boards must establish a clear risk appetite, enforce third-party audits, mandate incident reporting, and require supply-chain diligence across all AI vendors.

To operationalise AI governance, boards should systematically address the following core milestones:

Regulatory scan: Monitor the EU AI Act, US executive orders, and sector-specific rules in finance, health, and defence.

Risk taxonomy: Identify exposures in misuse, bias, data privacy, security, intellectual property, and geopolitical risk.

Governance structure: Form specialised board committees, establish AI ethics reviews, mandate regular red-team exercises, and define clear incident escalation paths.

Vendor diligence: Require audit rights, safety reports, model cards, explicit refusal policies, and data provenance guarantees before onboarding software.

Key metrics: Track audit pass rates, incident response times, human override frequency, and bias testing outcomes.

Accountability frameworks: Clearly define liability when AI informs high-stakes decisions in hiring, lending, healthcare, defence, or public services.

Keeping humanity at the heart of AI means choosing restraint, equity, and democratic oversight over unvetted speed and short-term profit. Technology should serve life, not jeopardise it.

For modern business, governance is not a brake on innovation. It is the prerequisite for durable trust, long-term value, and an enduring licence to operate.

Tan Sri Rainer Althoff is a strategic consultant on global digital transformation and a former CEO of Siemens Malaysia.