Google Expands Gemini’s Agentic Powers and Lengthens Chat Retention as AI Reshapes Security Concerns
Google is pushing its Gemini assistant further into autonomous, agent-style work while overhauling how long it holds onto users’ conversations, framing the longer retention as a safety and security measure for an era of increasingly capable AI. The changes arrived in the same week the company moved to dismiss reports of a Gmail outage, as U.S. officials linked a recent spate of service disruptions to the large-scale patching of systems, a trend partly driven by AI’s growing ability to uncover software vulnerabilities.
Gemini moves from chatting to doing
The most visible update is the wider rollout of Tasks, a mode in the Gemini app that turns the chatbot from an information tool into an active assistant. Where the standard Chat mode lets people converse with Google’s AI, Tasks allows Gemini to carry out practical errands on a user’s behalf, functioning much like a personal secretary.
Google says the feature can phone businesses and book appointments, complete purchases through Google Pay and sign into websites securely when prompted. In a blog post, Josh Woodward, Google’s VP of Gemini, said he had recently used it to call local car dealerships and schedule a service appointment for his family’s minivan, and also to analyse his electricity bill.
The capability has been evolving for months. It was first introduced to select testers in May as Gemini Spark, a round-the-clock agentic helper, before being extended to U.S. subscribers on the Google AI Ultra tier. Spark was later reworked into Task Mode. Google is now offering the feature free to the hundreds of millions of people using its new Gemini 4 Hydrogen model.
Activating it is straightforward. Users switch a toggle in the app from Chat to Tasks, and can also text or call Gemini to give it instructions.
Temporary chats now kept for around two months
Alongside the agentic expansion, Google is significantly lengthening how long it keeps temporary Gemini chats and conversations held with chat history switched off. The updated policy, rolling out globally in the Gemini app over the coming weeks, extends retention to roughly two months, up from the 72 hours stated in Google’s Gemini Apps Privacy Hub as recently as November 12, according to an archived version of the page.
That puts Google well ahead of its main rivals on retention. OpenAI says it may keep temporary ChatGPT chats for up to 30 days, while Anthropic retains incognito chats in Claude for 30 days by default. All of these companies may hold certain data for longer for legal or safety reasons, or to investigate policy violations.
Google did not explain what prompted the change or whether it would alter how conversations are accessed or reviewed. Instead, it pointed to a blog post stating that chats are retained longer so the company can respond to users, process feedback and protect its services and users. The refreshed Privacy Hub also cites potential system failures as a reason for keeping chats, and states that temporary conversations are not used to train AI models.
Chat history and AI training finally separated
The retention change comes as Google splits two controls that had previously been bundled together. Until now, a single Keep Activity setting governed both whether Gemini saved a user’s chat history and whether those conversations could be used to train Google’s generative AI models. Under the new arrangement, users can keep their conversations in their history without allowing them to be used for model training.
Google says both new settings will be off by default for anyone who had previously disabled Keep Activity. Those who had it switched on will keep their chat history enabled, but can now decide independently whether their conversations feed into AI training.
Reliability anxiety and the AI patching wave
Security and resilience were also on Google’s mind elsewhere. On Tuesday, the company posted on X that it was seeing no evidence of a broad Gmail disruption, directing users to its real-time status dashboard. The statement was aimed at outage reports circulating on the platform at the same time Apple was dealing with several confirmed service failures.
The Gmail concerns appear to have stemmed from a post by Downdetector, a crowdsourced site where users report problems themselves. Downdetector is not an official source for verifying outages, and its usefulness may decline if every minor glitch is reported as a full outage by non-technical users anxious about possible cyberattacks. While some users on X described Gmail problems, others reported none.
That Google felt the need to deny an outage at all reflects growing unease about the reliability of major digital services. Over the past week, Apple and Microsoft both suffered outages, an FAA network failure triggered a nationwide ground stop, and some banks and airlines also reported disruptions.
Director of National Intelligence Jay Clayton, speaking on Fox Business Network, said the U.S. was not facing coordinated cyberattacks and saw no signs of state-sponsored activity behind the incidents. He attributed the disruptions instead to businesses upgrading and patching their systems at a far larger scale than before, adding that some of those systems were not coming back online as expected in every case.
That patching surge may itself be a downstream effect of AI. Newer models have become increasingly adept at finding previously unknown vulnerabilities, leaving organisations racing to fix flaws faster than ever. Taken together, Google’s week illustrates the dual pressures facing the industry: handing AI agents more power to act for users, while managing the security, privacy and reliability consequences that increasingly capable AI brings with it.