
SSPAI Morning Brief: Google Expands Gemini API With Multimodal File Search and Advanced RAG Capabilities
Morning Brief
- DJI Unveils ROMO 2 Robot Vacuum Series
- Qwen Fully Integrates With Taobao
- Google Security Team Detects First Signs of AI-Assisted Hacking Activity
- Google Expands File Search Capabilities in Gemini API
- SoftBank to Produce Large Batteries for AI Data Centers at Former Sharp Factory
- Apple Acquires Color Science Company Patchflyer
- News Worth a Quick Look
DJI Unveils ROMO 2 Robot Vacuum Series
On May 11, DJI introduced the ROMO 2 series AI-powered robot vacuum and mop lineup. According to DJI, the ROMO 2 series features AI-powered cleaning capabilities that can automatically analyze various household cleaning scenarios and adjust cleaning logic based on different types of messes, including particles, heavy dust, and complex liquid spills. The device supports a 123° outward-extending body design, suction power of up to 36,000 Pa, and mechanical legs capable of crossing 4 cm single-layer obstacles and 8.5 cm double-layer obstacles. It also supports remote monitoring, voice control, and fast charging. The ROMO A2 (Advanced Edition) starts at RMB 5,499, while the ROMO P2 (Flagship Edition) starts at RMB 5,999. DJI also offers the standard ROMO S V2, starting at RMB 4,299.Source

Qwen Fully Integrates With Taobao
On May 11, Alibaba announced the full integration of Qwen with Taobao. Users only need to update the Qwen app to the latest version (6.9.1 or above) to complete product selection, comparison, and purchases directly from Taobao within the app. According to Alibaba, Qwen can leverage Taobao’s database of over 4 billion products and more than 20 years of real shopping scenario data to understand purchasing intent expressed in natural conversation. It is designed to accurately address three major shopping difficulties: “knowing what to buy but struggling with too many search conditions,” “knowing what to buy but unable to clearly describe the style or requirements,” and “having a clear usage scenario but not knowing what product to choose.” Meanwhile, the Taobao app has also launched the Qwen AI shopping assistant, supporting features such as AI virtual try-ons, AI-powered recommendations, and AI-assisted savings tools.Source
Google Security Team Detects First Signs of AI-Assisted Hacking Activity
On May 12, Google’s Threat Intelligence Group (GTIG) published a report titled “GTIG AI Threat Tracking: Adversaries Using Artificial Intelligence for Exploitation, Enhanced Operations, and Initial Access.” In the report, GTIG stated that while discovering and blocking a zero-day exploit attack, researchers identified signs of AI-assisted code generation within a Python script used by attackers for the first time.
Although the quality of AI-generated code remains inconsistent at this stage, the report noted that AI has significantly lowered the technical barrier for non-professional attackers to carry out large-scale cyber intrusions, dramatically shortening the weaponization process for zero-day vulnerabilities. The report also warned that attackers are using large language models to generate highly deceptive phishing emails and are crafting specialized prompts to manipulate enterprise AI assistants into leaking sensitive information or performing unauthorized actions.
GTIG recommends that security teams adopt an “AI versus AI” strategy by deploying automated analysis and detection models capable of identifying AI-generated malicious payloads and abnormal traffic in real time. At the same time, enterprises are encouraged to strengthen employee training focused on identifying AI-generated content.Source
Google Expands File Search Capabilities in Gemini API
Google recently announced an expansion of file search capabilities within the Google Gemini API, bringing developers more comprehensive multimodal Retrieval-Augmented Generation (RAG) functionality. The core updates include support for mixed image-and-text retrieval, custom metadata filtering, and page-level citations, improving accessibility and accuracy for enterprise knowledge bases, document Q&A systems, and AI agents.
According to Google’s official blog, the updated file search system is no longer limited to traditional text vector search. Instead, it is built on Gemini Embedding 2’s unified multimodal embedding capabilities, allowing the system to simultaneously understand visual and textual content across images, PDFs, and documents. Developers no longer need to build complex vector databases, embedding pipelines, or document chunking systems themselves, and can instead complete full RAG workflows directly through the Gemini API.
The custom metadata filtering feature allows developers to add metadata such as tags, categories, timestamps, and departments to uploaded files, enabling more accurate and efficient filtering during retrieval. Meanwhile, the new page-level citation functionality allows Gemini to explicitly reference the exact document page where information originates, rather than vaguely citing the entire file. All new features are now officially available.Source
SoftBank to Produce Large Batteries for AI Data Centers at Former Sharp Factory
SoftBank’s mobile business subsidiary recently announced plans to build a large-scale battery production line at a former Sharp factory in Sakai, Osaka, aiming to provide power infrastructure for its rapidly expanding AI data center operations. The factory is scheduled to begin mass production during fiscal year 2026, with a target annual production capacity of 1 gigawatt-hour (GWh).
The site previously served as one of Sharp’s well-known LCD panel manufacturing centers. Through partnerships with South Korean companies Cosmos Lab and DeltaX, SoftBank will initially produce battery types including lithium iron phosphate (LFP) batteries, while planning to introduce non-flammable zinc-halide battery technology in 2027 that does not rely on China’s rare metal supply chain.Source
Apple Acquires Color Science Company Patchflyer
A recent acquisition disclosure filing from the European Union has confirmed that Apple completed its acquisition of Patchflyer in January 2026. The company was founded by color science expert Jonathan Ochmann, whose core product was the online color grading tool Color.io, widely favored by photographers and filmmakers. The service announced it would shut down late last year, and its founder has since joined Apple’s imaging team. The filing also revealed that Apple acquired computer vision startup PromptAI during the same period. PromptAI previously developed an app called Seemour, focused on using AI technology to enhance recognition capabilities for home security cameras. Source
News Worth a Quick Look
- Apple has released version 26.5 updates across its operating systems, introducing multiple new features including end-to-end encrypted RCS messaging between iOS and Android devices. At the same time, Apple also pushed security-focused updates for systems running versions 15 through 17.
- Kuaishou plans to spin off Kling AI and pursue a 2027 IPO at a reported valuation of $20 billion. Source
- Linux 7.2 will discontinue support for i586/i686 CPUs lacking timestamp counter (TSC) instructions, including processors such as the AMD K5. Source
- TikTok has announced the launch of TikTok Ad-Free, an ad-free subscription service in the UK priced at £3.99 per month. Source
- According to Neowin, some Dell users have recently experienced frequent blue screen crashes and reboots following software updates. Investigations found that the issue was not caused by Windows 11 itself, but by Dell’s preinstalled SupportAssist software triggering abnormal behavior in the WinDbg debugger, ultimately causing kernel-level shutdowns. Users experiencing similar problems can reportedly resolve the issue simply by uninstalling SupportAssist. Notably, the same Dell software caused a similar incident back in December 2024. Source
- British singer Dua Lipa has filed a $15 million lawsuit against Samsung over infringement claims. Since last year, Samsung has allegedly used backstage photos from her 2024 tour on television packaging without authorization, while repeatedly ignoring requests from Dua Lipa’s team to replace the packaging. Evidence submitted by her legal team reportedly includes multiple social media comments from consumers who purchased Samsung TVs specifically because the packaging featured Dua Lipa. Source

- Microsoft is reportedly preparing to introduce a new performance enhancement feature for Windows 11 called “Low Latency Profile.” The feature aims to significantly reduce response times by temporarily boosting CPU frequency to maximum levels during app launches and system interactions. Some users in technical communities criticized the approach on social media, arguing that Microsoft should focus on optimizing underlying code instead of taking “shortcuts.” Microsoft Vice President Scott Hanselman publicly responded on X on May 10, explaining that this is a standard practice in modern operating systems rather than “cheating.” He noted that systems including macOS, Linux, and even smartphone operating systems all temporarily raise CPU frequencies to ensure smoother interactions, bluntly adding: “Apple does this too, and you all seem to love it.” Source


Leave a Reply