Apple ’s conclusion to delay the bankroll out of the vamp Siri assistant features has led to plenty of speculation over why the company is n’t yet able-bodied to deliver the Apple Intelligence - establish upgrade .
While Apple itself admitted its ‘ taking longer than we think ‘ to get the smarter and more contextually aware Siri ready for prime clip , other observers have unlike hypothesis on why the feature designed for thebest iPhonemodels is experiencing developing setbacks .
Bloomberg ’s Mark Gurman of late describe the party is struggling with consistency for these features , which include the power for apps to take activity on the user ’s behalf in what ’s being call up “ agentic ” AI .
However one developer thinks , current reliability aside , the key issue Apple is grappling with is the surety vulnerability inherent to the unexampled technology .
Developer Simon Willison ( via9to5Mac ) reckons the internal concern may be over something called prompt injectant attacks , which are an issue that occur when malicious actors are able to trick an AI that the request is coming from a drug user .
On hiswebsiteWillison wrote : “ These new Apple Intelligence features involve Siri respond to request to get at selective information in app and then performing legal action on the user ’s behalf .
“ This is the worst possible combining for prompt injectant attacks ! Any clock time an LLM - based system has access to private data , pecker it can call , and pic to potentially malicious instruction ( like emails and schoolbook messages from untrusted stranger ) there ’s a significant peril that an assailant might vitiate those tool and use them to damage or exfiltrating a user ’s datum . ”
unluckily , this is n’t something Apple will easily be able-bodied to get around because , at present , large voice communication models are inherently susceptible to these kind of onset . As Daring Fireball ’s John Gruberputs it , Apple would have to make “ groundbreaking ” development to overcome this issue .
He wrote : “ They have to solve a vexing problem — as yet unsolved by OpenAI , Google , or any other result AI laboratory — to birth what they ’ve already promise . ”
Opinion
Apple continually stake its reputation on being a privacy - first caller and has developed trust among the user fundament for that cause . We saw very latterly that Apple is not all talk on this front , pulling an encoding feature in the UK because the government had demanded a back door into it .
A buggy liberation for this new technology would be largely tolerated by users . However , something that compromise their on-line security would have huge , huge consequences .
It ’s a hard fleck for Apple because it can not afford to be left behind in the AI race , but that may mean take chance it is not well-heeled with .
