A couple of stories from the last few weeks show how keen people (and their trillion dollar companies) are to comment on AI and how it might be best to take a step back before getting involved.
Apple is reportedly building its own AI servers, but at WWDC recently, CEO Tim Cook announced that iOS 18, the iPhone operating system that will be released later this year, will pair Apple’s Siri with OpenAI’s ChatGPT to provide extensive AI capabilities. Apple tweeted, ‘Introducing Apple Intelligence for iPhone, iPad and Mac’ and Cook replied, ‘It’s personal, powerful and private – and it’s integrated into the apps you rely on every day’.
Elon Musk (who has history with OpenAI, a company he co-founded in 2015 before leaving without explanation in 2018 and can rarely resist taking a dig at) immediately took to X, telling Cook ‘Don’t want it. Either stop this creepy software, or all Apple devices will be banned from the premises of my companies’.
Musk’s comments were based on the assumption that ChatGPT would be integrated at OS level on Apple devices, meaning the potential would exist for a large amount of personal data to be passed from device to OpenAI. On their website, OpenAI posted that privacy protections were built in when accessing ChatGPT through Siri and any writing tools, that requests are not stored by OpenAI, and that IP addresses are obscured. A community note on X, which was subsequently removed, pointed out that Apple Intelligence does not give personal data to OpenAI.
Clearly, linking Siri to ChatGPT has the potential to hugely increase AI capabilities, and the data passing between two platforms means both parties need to be particularly aware of the hugely increased security concerns.
Less personal but displaying a similar need to foist an AI solution on the public before considering the security angle was Microsoft’s announcement to include and then turn off by default Microsoft Recall, a feature from their new range of tablets and laptops.
The new Surface range includes processors with neural processing units which boost AI processing. These Copilot+ PCs have sufficient power that Microsoft could include Recall, a tool that could record the Windows desktop any time the desktop content changed. So, it would be possible to record all information that ever appeared on your laptop or tablet, whether in a note, on a web page from your bank or WhatsApp – absolutely anything.
That data was captured as a screenshot, run through optical character recognition software, and saved in a searchable database. By default, 25GB of data was allocated to the database, which Microsoft estimated should be good for three months of snapshots.
The database is stored locally (so there’s no security risk of a cloud service being compromised and thousands of users’ data being accessed or of data in transit being intercepted), and the data is encrypted, giving some security against the device being stolen and the database being read.
However, as security expert Kevin Beaumont pointed out, malware designed to steal information after being inadvertently installed on PCs is hardly novel, and stolen credentials already number in the billions. Now, that could be scaled globally with three months of data available.
AI brings enormous benefits to productivity, but it also significantly increases the security risk personally and professionally, and it is vital to consider the risks before implementing the benefits.