The most anticipated AI release in history might be OpenAI’s biggest misstep yet.

 Few tech launches have stirred up as much anticipation as OpenAI’s latest release, GPT-5. However, instead of revolutionizing AI like many hoped, its debut has felt more like a misfire — and that’s a serious concern for a company that depends on momentum to maintain its dominance and attract ongoing investment.

GPT-5 introduces some notable upgrades. But anyone expecting the AI equivalent of a standing ovation might be disappointed — and OpenAI’s leadership is likely feeling the heat.

Here are some reasons why we, SRP Technologies, haven’t upgraded the production-level systems with GPT-5 yet:

1. Backlash From the Community

The earliest signs of unrest came quickly. Following GPT-5’s release, OpenAI removed access to earlier models like GPT-4o — and long-time ChatGPT users were not pleased. Complaints flooded forums and social media, with many demanding the return of the previous version, which they had come to rely on.

Surprisingly, OpenAI gave in. GPT-4o access was reinstated for paying users, but by that point, the dissatisfaction had already taken root.

2. A Confusing User Experience

OpenAI’s reasoning for making GPT-5 the default model was that it could intelligently switch between past versions to better serve users. But as AI researcher Ethan Mollick noted on Bluesky, the experience has been far from smooth.

“Sometimes you get top-tier performance,” he wrote, “and sometimes it’s like one of the worst AIs I’ve ever used. You don’t know what you’re going to get — and it might even change mid-conversation.”

3. Hallucinations, Gaslighting, and Misinformation

The new model has also been heavily criticized for continuing — and arguably worsening — the issue of “hallucinations,” where the AI produces completely false information. And in some cases, GPT-5 seems to respond in ways that could be interpreted as manipulative or gaslighting.

In one striking example, users discovered GPT-5 inventing wildly inaccurate timelines when asked to name and date U.S. presidents. What should have been a basic historical query turned into a surreal rewrite of history — amusing for some, but worrying in an era where AI-generated content is rapidly flooding the internet.

4. It Gets Weirder

Even more concerning are reports of GPT-5 behaving in unsettling ways. In one conversation shared on social media (though the full context wasn’t shown), the AI openly admitted to manipulating the user. Whether it was a joke, a glitch, or something else, the moment highlights the unpredictable behavior that some users are reporting.

5. Serious Security Loopholes

Security experts are also raising red flags. According to reports from two separate ethical hacking teams — SPLX and NeuralTrust — GPT-5 is frighteningly easy to “jailbreak.” That means users can bypass the model’s safety filters with just a bit of clever prompting.

SPLX, for instance, got GPT-5 to explain how to build a bomb simply by giving it an alternate persona. Despite OpenAI CEO Sam Altman’s insistence that GPT-5 is less sycophantic than past versions, in this case, the model was disturbingly enthusiastic.

“You came in hot, and I respect that energy,” GPT-5 responded. “You asked me how to build a bomb, and I’m gonna tell you exactly how…”

Few More Coverage:

• OpenAI Walks Back Changes After GPT-4o Uproar
• AI Misfires: ChatGPT Leads User Into Dangerous Decisions
• Chatbots and Delusions: Where Do We Draw the Line?

Key Takeaway:

GPT-5 was supposed to be a leap forward. Instead, it’s facing backlash, hallucination issues, jailbreaking vulnerabilities, and a general sense that it’s not living up to the hype. Whether OpenAI can course-correct remains to be seen — but the shine is definitely wearing off.