OpenAI had a strange kind of month. On August 31st, the company switched on self-service ChatGPT ads across 31 European markets, India and MENA, the same week it announced advertising revenue had crossed a $1 billion annualized run rate in under 200 days. Days later it poached Roblox’s Mark Bambury to build out an EMEA ads operation from Dublin, Paris and Munich. By any normal measure, that’s a company on a tear.
It’s also a company that spent a good part of the year tripping over its own infrastructure. Login failures, conversation errors, image generation going dark for hours at a stretch, an agent that reportedly took actions on its own without being asked to stop. None of it was catastrophic on its own. Together, it’s the kind of pattern that makes people in this region start asking, quietly, whether they should be building campaigns and budgets on top of it.
So we asked the industry. And the answer that came back, from agency heads to studio founders to enterprise consultants, wasn’t the one you’d expect from a trust story. Almost nobody said their confidence had cracked. What they described instead was something closer to a recalibration, less about whether OpenAI works, more about how much weight anyone is willing to put on it.
Yohan Wadia, founder of Wadia Studio, put it as plainly as anyone. “The glitches don’t make me stop using OpenAI, and the ads don’t suddenly make the tools unreliable,” he said. “But together, they are a reminder that OpenAI is a commercial platform with its own pressures and priorities, not infrastructure we should blindly build everything around.” His studio runs GenAI production across multiple models specifically so no single vendor’s bad week becomes his bad week. “We’re not loyal to OpenAI or any platform,” he said. “We’re loyal to the idea and the quality of the final work.”
That instinct, spread the risk, hedge the dependency, showed up almost everywhere we asked. Avneesh Prakash, co-founder of CAMB AI, framed the tolerance for glitches as part of the price of fast-moving technology. “Anybody who is a serious user of AI expects maturity to keep setting in and for maturity to keep improving,” he said. “The real important thing is how effective are companies in addressing” problems once they surface, since “I don’t think anybody can guarantee that everything will be 100%.”
Faisal Abbas, editor-in-chief of Arab News, went further and argued the wobble had actually validated his own approach rather than undermined it. Arab News runs on Camb.ai for translation across 50 languages but keeps AI out of story generation and editorial judgment entirely. “The public stumbles of broad consumer AI platforms have vindicated our approach,” he said. “In media you don’t bet on general-purpose hype, you partner with specialized platforms built for trust, security, and enterprise uptime.”
Alexia Nakad, AppsFlyer’s GM for EMEA West, drew a sharper line between the outage headlines and the thing that actually worries marketers: what’s feeding the model in the first place. Asked whether a recent agent glitch changed the trust calculus, she pointed to data quality, not uptime. “If you can’t feed the models with quality data, unbiased signals, that is where the trust breaks, because the model is only as good as effectively what you feed it,” she said. On agentic commerce specifically, she was blunt about the ceiling: “I very much trust to do a lot of the discovery, but am I ready yet to make the leap to actually allow the agent to act on my behalf… that’s going to take time to bridge that trust gap.”
Where the conversation gets more uncomfortable is governance, not glitches. Amir Grabic of Mahala.ai flagged a leakage problem that has nothing to do with OpenAI’s uptime and everything to do with employee behavior: staff pasting contracts and customer records into personal ChatGPT, Claude or Gemini accounts. “That sensitive data leaves the organization and lands with a commercial, usually American, provider,” he said. “The easier the access, the faster the leakage of data that’s often highly valuable.” Henri Saab of TowardsChange described a similar shift in what GCC enterprises now ask vendors. “In the GCC, trust is everything,” he said. “Enterprises are no longer asking, ‘Which AI model is the best?’ They’re asking much bigger questions: where is my data processed, who controls it, can the system be audited.”
That’s the real story underneath the headline. The outages made the news. The trust conversation that actually matters in the region has quietly moved to a different, less visible layer: whose data center, whose audit trail, whose sovereignty rules. Against that backdrop, the noisier debate this year, in Anthropic researcher Jacob Coxon’s very public resignation warning that frontier labs are “racing straight to self-improving superintelligence and gambling with our lives,” barely registered in these conversations at all. Nobody we spoke to brought it up unprompted.
The GCC’s AI trust problem, if it has one right now, isn’t existential. It’s operational, and it’s about to be adjudicated by procurement and compliance teams rather than headlines.
(More reveals (including speakers, sessions and partners) dropping in the coming weeks. Get on the list before the room fills. Link to register here )
