By CultureBanx Team
- Salesforce AI policy executive wants people to treat AI like an intern: powerful and productive, but never a substitute for human judgment or accountability
- A Pew Research Center study found 55% of both U.S. adults and AI experts surveyed were highly concerned about bias in AI decision-making
Artificial intelligence is quickly graduating from chatbot to co-worker. As autonomous AI agents gain the ability to complete tasks, make recommendations and take actions on behalf of users, one question is becoming impossible to ignore: Who is actually in charge? For Dr. Rachel Gillum, Vice President of Ethical & Humane Use of Technology at Salesforce, the answer should still be us.
Why This Matters: Globally, the debate over human oversight is about much more than whether ChatGPT gets an answer wrong. AI is increasingly influencing systems tied to employment, credit, healthcare, education and government services, areas where inequities already exist.
“It is so critical that we don’t delegate our whole brains and judgment to these tools,” Gillum told CultureBanx during its Reimagining Our Tech Agency series. Her advice for navigating this new reality is surprisingly simple: treat AI like your intern.
An intern might bring energy, technical skills and new ideas, Gillum explained, but may lack the context, judgment and common sense necessary to make the final call. “You are the manager, you are the person responsible,” she said. “You really need to lean in and make decisions about, again, what you delegate and then what you deliver as the final product as the manager in charge.”
That distinction becomes much more important as AI moves beyond generating emails and meeting summaries into decisions that can materially affect people’s lives.
AI oversight: AI-driven workplace disruption could disproportionately hit Black workers. McKinsey research on generative AI and Black communities found that about 24% of Black workers were employed in occupations with more than 75% automation potential, compared with 20% of White workers.
The trust gap is already visible. A 2025 Pew Research Center study found 55% of both U.S. adults and AI experts surveyed were highly concerned about bias in AI decision-making. Experts were also considerably less likely to believe Black and Hispanic perspectives were well represented in AI design than White perspectives.
Recognizing those risks isn’t enough. “We cannot just set it and forget it, be asleep at the wheel,” she told CultureBanx. “We need citizens to lean in and to push back when there is a concern.”
Building Guardrails: The data suggests the concerns around AI aren’t hypothetical. Stanford University’s 2026 AI Index report found 362 documented AI incidents in 2025, up from 233 in 2024. At the same time, Stanford found that responsible AI testing isn’t keeping pace with the rapid development and deployment of increasingly capable systems.
Transparency is another major problem. The AI Index report also found the average Foundation Model Transparency Index score among evaluated developers fell from 58 in 2024 to 40 in 2025. That means the AI economy may be accelerating at precisely the moment when understanding how some powerful systems are trained, tested and monitored remains difficult.
Responsible AI is also broader than racial fairness. The National Institute of Standards and Technology’s AI Risk Management Framework was developed to help organizations manage AI risks affecting individuals, businesses and society. NIST identifies trustworthy AI characteristics that include systems being valid and reliable, safe, secure and resilient, accountable and transparent, explainable, privacy enhanced and fair, with harmful bias managed.
Critically, NIST says organizations should consider those characteristics throughout an AI system’s lifecycle, from initial design and development through deployment, use, testing and evaluation. That closely mirrors Gillum’s argument that responsible technology can’t be something corporations bolt onto a product after development.
“One way we approach it, first of all, is implementing these guardrails and work by design from the beginning, so it’s not tacked on at the end,” she said.
Challenging The Algorithm: Ethical AI is also increasingly becoming a question of consumer rights. Colorado, for example, has enacted protections governing automated decision making technology used in consequential decisions. Under provisions scheduled to take effect in January 2027, consumers will have rights that include requesting and correcting inaccurate personal data used by automated decision-making systems, according to the Colorado Attorney General.
That’s important because accountability becomes much harder when people don’t know an algorithm was involved, can’t understand what information influenced it or have no meaningful way to challenge a bad outcome. Gillum argues that communities historically left outside technology’s decision making rooms therefore need more than protection from AI, they need actual influence over it.
“They need to have a central role because these tools are being built for everyone in society,” she said. “Everyone should be taking a seat at the table.”
CBX Vibe: “Friend’s Again” Baby Rose & Leon Thomas
