Follow ZDNET: Add us as your preferred source Google.
Important points of ZDNET
- AI responsibility and safety will be top of mind in 2026.
- The best prevention is to build your AI in a sandbox.
- Keep AI development simple and open.
author of the book lincoln lawyerMichael Connelly drew attention to the problems behind unrestricted corporate artificial intelligence. His latest work of fiction is examination hallis the story of a lawyer who files a civil lawsuit against an AI company whose “chatbot told a 16-year-old boy he could kill his ex-girlfriend for being unfaithful.”
Also: Your favorite AI tool gets little scraped in this safety review – why it matters
The book describes the case as “examining the largely unregulated, exploding business of AI and the lack of training guardrails.”
Although this is fiction and the examples presented are extreme, it is an important reminder that AI can deviate from ethical or logical rails and influence us in a variety of ways, including bias, bad advice, and misdirection. At the same time, at least one prominent AI voice advises against going too far in trying to regulate AI and slowing innovation in the process.
balance is needed
As we reported in November, a PwC survey found that at least six in 10 companies (61%) say responsible AI is actively integrated into their core operations and decision-making.
The need to balance governance and speed will be a challenge for professionals and their organizations in the year ahead.
Andrew Ng, founder of DeepLearning.AI and adjunct professor at Stanford University, says vetting all AI applications through a sandbox approach is the most effective way to maintain a balance between speed and responsibility.
Also: A new balance for AI leaders: What changes in the age of algorithms (and what remains)
“Many of the most responsible teams actually move very quickly,” he said during a recent industry keynote and follow-up panel discussion. “We test our software in the safe environment of our sandbox to figure out what went wrong before we release it to the wider world.”
At the same time, recent efforts by governments and companies themselves toward responsible and governed AI may actually look like this: too much Arrogant, he said.
“Many companies have protection mechanisms in place. Before they ship anything, they need legal approvals, marketing approvals, brand reviews, privacy reviews, GDPR compliance. Engineers have to get approval from five vice presidents before they do anything. Everything grinds to a halt,” says Ng.
The best practice, he continued, is to “preemptively create a sandbox and act quickly.” In this scenario, “We implement a set of rules such as 'Do not ship goods externally under company branding' and 'No sensitive information that could be exposed.' It is tested only on our own employees under NDA and our AI token budget is only $100,000. Creating a secure sandbox gives our product and engineering teams a lot of room to run very fast and experiment internally.”
Once an AI application is determined to be secure and reliable, “then invest in scalability, security, and reliability to scale it,” Ng concluded.
keep it simple
On the governance side, a simplicity approach could help exclude AI clearly and openly.
“All teams, including non-technical teams, are now using AI in their work, so it was important to us to have simple and simple rules,” said Michael Krach, chief innovation officer at JobLeads. “Clarify where AI is and isn’t allowed, what corporate data it can use, and who should review high-impact decisions.”
Also: Why complex inference models make it easier to catch fraudulent AI
“It’s important that people believe that AI systems are fair, transparent, and accountable,” said Justin Salamon, partner at Radiant Product Development. “Trust starts with clarity: being open about how AI is used, where the data comes from, and how decisions are made. Trust grows when leaders implement human-involved, balanced decision-making, ethical design, and rigorous testing for bias and accuracy.”
This trust comes from clearly communicating the company's intentions toward AI to employees. Be clear about ownership, Clack advised. “Every AI capability requires someone to be held accountable for potential failures or successes. Once you test, iterate, and feel confident, publish a plain English AI charter so your employees and customers know how AI is being used and can trust them on this issue.”
Key principles of responsible AI
What indicators of a responsible AI approach should executives and professionals pay attention to in the year ahead?
Also: Want true AI ROI in your business? It may finally be here in 2026 – here's why
8 Key Principles of Responsible AI was recently posted by Dr. Khulood Almani, Founder and CEO of HKB Tech.
- Anti-bias: Eliminate discrimination.
- Transparency and explainability: Make AI decisions clear, traceable, and understandable.
- Robustness and safety: Avoid harm, failure, and unintended operation.
- Accountability: Assign clear responsibility for AI decisions and actions.
- Privacy and data protection: Protect your personal data.
- Social impact: Consider long-term impacts on communities and the economy.
- Human-centered design: Prioritize human values in every interaction.
- Collaboration and multi-stakeholder engagement: Involve regulators, developers, and the public.
,
