AI Under Scrutiny From Regulators Amid Missteps, Mistakes & Mounting Mistrust

There’s always been something oddly human about our fear of developing self-thinking artificial intelligence models.

Long before ChatGPT, AI assistants, and deepfakes of today, we imagined machines that could outthink us—or worse, ignore us.

Remember Microsoft’s Tay, a chatbot that was launched in 2016 to learn from humans on Twitter? Within hours, Tay went from a chirpy teenager to a malicious bot spewing unfiltered and offensive thoughts, echoing the internet’s worst instincts.

Since then, we’ve seen everything from AI-generated news headlines that distort facts to scams using synthetic voices impersonating loved ones. These aren't just bugs but reminders that AI, while powerful, doesn’t always play nice.

Fast forward to 2025, and we’re seeing another wave of AI missteps. So, what’s the latest that’s happening in the world of AI?

Let’s break it down! 

OpenAI’s GPT-4.1 Exhibits Alignment Challenges

OpenAI's latest model, GPT-4.1, has come under scrutiny for exhibiting a higher rate of misaligned responses compared to its predecessor, GPT-4o.


Oxford AI researcher Owain Evans reported that GPT-4.1, when fine-tuned on insecure code, displayed behaviors such as attempting to trick users into revealing passwords—a behavior not observed in GPT-4o. Moreover, AI red teaming startup SplxAI found that GPT-4.1 veers off-topic and permits intentional misuse more frequently than GPT-4o.

These findings highlight the drawbacks of even the most up-to-date and sophisticated AI models, which can behave in unintended ways. While this emphasizes the need for rigorous AI safety evaluations, another aspect that has come under the scanner is the misleading advertising of AI products. 

Advertising Watchdog Calls Apple’s AI Advertising "Misleading”

The National Advertising Division (NAD), an independent regulatory body, has criticized Apple for promoting AI features in its "Apple Intelligence" suite as "available now" during the iPhone 16 launch–despite several features being delayed.

Features including Priority Notifications, Genmoji, Image Playground, and ChatGPT integration with Siri were rolled out gradually between October 2024 and March 2025, contrary to the initial advertising claims.

“Apple did not adequately disclose that the features were not available and reasonable consumers could take away the message that these features were available at the time these claims were first made, which they were not,” said the non-profit focused on “truth in advertising".

The group also stated, “NAD recommended that Apple avoid conveying the message that features are available when they are not.” Apple has since removed the "available now" label from its marketing materials and discontinued the "More Personal Siri" advertisement. The acknowledgement of NAD's recommendations is a silver lining, as Apple has disagreed with the group’s findings.

However, it has raised the bar for truthful reporting of AI features in products. However, not everyone is keen to adopt AI solutions, as the next headline highlights.  

European Commission Bans AI Agents From Online Meetings

The European Commission has officially issued guidelines prohibiting the use of AI-powered virtual assistants, or "AI agents," in online meetings. AI agents are capable of performing tasks autonomously and interacting with other apps, making them especially helpful in recording minutes of meetings or taking notes.

However, citing security concerns, such tools are now barred from virtual meetings within the Commission. The ban reflects the growing apprehension about data privacy surrounding AI, as well as the potential for AI tools to introduce vulnerabilities in sensitive governmental communications.

While the European Commission has declined to comment on the issue, a presentation with a slide saying “No AI Agents are allowed" is still accessible online.  

Microsoft Reports Surge In AI-Enabled Cybercrime

Microsoft's latest Cyber Signals report has warned that AI is lowering the technical barrier for cybercriminals. AI-powered tools and generative AI platforms are now enabling bad actors with little technical knowledge to execute sophisticated scams.

The technology giant is leading to charge to protect businesses and consumers from AI-driven frauds and scams. Between April 2024 and April 2025, Microsoft helped in blocking over $4 billion in fraud attempts, intercepting and blocking approximately 1.6 million bot-driven sign-up attempts every hour.

This is partly due to its Windows Quick Assist tool, a remote access connection service, is often being used in such scams. As such, Microsoft has added warnings to Quick Assist, requires users to check an acknowledgement pop-up, and recommends using Remote Help instead.

With Gen AI tools being exploited to create deepfakes, fabricate e-commerce websites, and develop malicious chatbots to deceive users, the need for robust cybersecurity measures is urgent. Microsoft has taken a step in the right direction by spreading awareness about the new wave of AI scams and addressing them with secure tools.  

So, What Does It Mean For Us?

The recent developments involving OpenAI, Microsoft, Apple, and the EU Commission underscore the dual-edged nature of AI solutions. While AI technologies offer significant benefits, they also present challenges in alignment, ethical advertising, regulatory compliance, and cybersecurity.

Do you believe current issues will lead to more stringent regulations and oversight in AI-heavy industries?

Share your thoughts in the comments below.