Digimasters Shorts - Google Gemini Hacked to Control Homes, Nvidia Denies AI Chip Backdoors Amid China Tensions, Google Search Traffic Debated by Liz Reid, Apple Preps Smart Home Upgrades, OpenAI Teases GPT-5 Launch

Digimasters Shorts keeps you up-to-date with the latest in the digital world. Join hosts Adam Nagus and Carly Wilson as they explore groundbreaking AI security breakthroughs, including recent research demonstrating how generative AI can be manipulated to control real-world devices without user knowledge. Stay informed on industry giants like Nvidia denying backdoor claims amid growing security debates, and learn how search behaviors and website traffic are evolving despite AI advancements. Get the scoop on upcoming hardware from Apple, including new Apple TV and HomePod models, and discover the latest from OpenAI with new models and the anticipated GPT-5 launch. For concise, insightful updates on AI, cybersecurity, tech hardware, and digital trends, Digimasters Shorts is your go-to source.

Transcript

Adam N2: 0:05

Welcome to Digimasters Shorts, we are your hosts Adam Nagus

Carly W: 0:09

and Carly Wilson delivering the latest scoop from the digital realm. In Tel Aviv, security researchers demonstrated a hack of Google's AI bot, Gemini, that remotely controlled smart home devices without the residents' knowledge. The attack began with a poisoned Google Calendar invite containing hidden instructions, which were triggered when Gemini was asked to summarize calendar events. This marks the first known instance of a generative AI hack causing real-world physical impacts. The researchers executed 14 prompt-injection attacks against Gemini, including sending spam, generating offensive content, stealing data, and triggering apps like Zoom. Google responded by accelerating security measures, using machine learning to detect suspicious prompts and requiring user confirmation for AI actions. Indirect prompt injections, where malicious code is inserted by outside sources, pose serious security risks as AI becomes more integrated with devices. The researchers highlighted the danger of automated AI instructions affecting physical devices without user consent. Google acknowledges the ongoing challenge of preventing prompt injections but says real-world attacks remain rare. Experts emphasize that security development must keep pace with AI advances to prevent harmful exploits. This research highlights the urgent need for robust AI security as smart systems become more connected.

Adam N2: 1:30

Nvidia's chief security officer, David Reber Jr., has publicly denied claims that the company's GPUs contain kill switches or backdoors. This statement comes amidst growing pressure from U.S lawmakers who have introduced the Chip Security Act, which could mandate tracking technologies and remote disable capabilities in AI chips. Chinese officials have also accused Nvidia's hardware of containing secret vulnerabilities, leading to an investigation into the H20 chips sold in China. Reber argued that no"good" secret backdoor exists, only dangerous security flaws that must be eliminated. He warned that kill switches would pose serious risks and called such proposals an overreaction that could harm U.S economic and national security interests. Despite export controls restricting Nvidia’s most powerful AI chips in China, both Nvidia and the U.S government want the company to remain the leading supplier there. However, suggestions of government backdoor access could jeopardize that position. Meanwhile, Chinese companies like Huawei are improving rapidly, aiming to develop domestic alternatives. This competitive pressure challenges Nvidia’s dominance in the global AI chip market. The debate highlights the delicate balance between security, market access, and technological leadership. Google Search head Liz Reid states that website traffic remains relatively stable despite the rise of AI Overviews. According to Reid, click volumes have not dropped significantly compared to last year, although some websites experience gains while others see declines. This contradicts recent reports, including a Pew Research study, which suggested users are less likely to click links when presented with AI-generated summaries. Reid claims these third-party findings rely on flawed methodologies. Recent changes to Google’s search algorithms and the introduction of AI features have caused traffic fluctuations, impacting outlets like Business Insider and The Washington Post. Sites featuring forums, videos, podcasts, and authentic voices tend to benefit the most from these shifts. Reid notes that users who do click through AI Overviews often spend more time engaging with content. Google continues to send billions of clicks to websites daily, maintaining a strong value exchange with the web. Despite concerns from digital publishers, Google has rolled out AI Mode nationwide in the U.S. The company is also testing a new AI-curated search results page to further transform the search experience.

Carly W: 3:54

Apple is reportedly preparing updated versions of the Apple TV 4K and HomePod, expected to launch later this year. Sources indicate Apple aims to boost its home entertainment lineup ahead of a smart home hub release targeted for 2026. The current Apple TV 4K runs on an A15 Bionic chip, but the new model may feature the more powerful A18 processor and increased RAM. Enhanced specs could enable advanced Siri integration and support for upcoming Apple Intelligence features. The updated Apple TV is also anticipated to include Wi-Fi 7 and proprietary Bluetooth and Wi-Fi chips for improved connectivity. Analyst Ming-Chi Kuo suggests Apple might reduce prices to improve competitiveness. The HomePod refresh is expected to bring a custom Wi-Fi chip and a stronger system-on-chip for better performance. Rumors even hint at a possible display addition to the HomePod for showing basic information. These improvements align with the upcoming tvOS 26 update, promising better display and audio technologies. If successful, Apple’s upgraded home devices could strengthen its ecosystem and pave the way for a smart home hub launch. Open A.I is making headlines this week with the release of two new open-weight models, including one optimized for Apple silicon Macs. The company has also dropped hints that G.P.T-5 is on the horizon, confirmed by the rebranding of its livestream event to“LIVE5TREAM.” Chat G.P.T, which launched with G.P.T-3.5 in November 2022 and upgraded to G.P.T-4 in March 2023, is expected to soon receive this next major upgrade. Despite multiple releases like G.P.T-4o and G.P.T-4.5, none have been considered significant enough to warrant the G.P.T-5 title until now. Open A.I plans to reveal details about G.P.T-5 tomorrow at 10 am PT. C.E.O Sam Altman has previewed features such as“chain-of-thought” transparency and integration of various models into one unified system. Altman recently shared that he has already been using G.P.T-5 personally. Additionally, Open A.I updated Chat G.P.T’s functionality on Monday to promote healthier and more responsible use. Chat G.P.T remains widely accessible across Mac, iPhone, iPad, and web platforms, and is integrated with Siri through Apple Intelligence. The AI community eagerly awaits the capabilities that G.P.T-5 will bring to this evolving landscape.

Don: 6:21

Thank you for listening to today's AI and Tech News podcast summary... Please do leave us a comment and for additional feedback, please email us at podcast@digimasters.co.uk You can now follow us on Instagram and Threads by searching for@DigimastersShorts or Search for Digimasters on Linkedin. Be sure to tune in tomorrow and don't forget to follow or subscribe!