Google Gemini Smartglasses Start at $395 as Privacy Concerns Remain

Key Takeaways

• AI-agent response plans should account for mistakes as well as attacks, identify responsible decision-makers and preserve records of agent identities, handoffs and tool actions.
• Agent offboarding requires revoking credentials and tracing service accounts, shared tokens, scheduled jobs and delegated work; preserve the context behind prior decisions.
• Google’s Gemini-powered smartglasses are priced from $395, with comfort and privacy remaining key concerns.
• UK GDP grew 0.7% in Q3 2026; computer programming and consultancy added 0.15 percentage points, but the ONS did not separate AI-related growth.
• In a survey of 150 Michigan small-business respondents, 71% use AI and 71% expect a positive impact; 62% also see risks, especially data security and job displacement.
• Business owners interviewed by the Daily Nebraskan described AI as useful for marketing efficiency and trend analysis, while stressing the value of human creativity and an authentic brand.
• Cornell plans five AI and data science summits covering social good, healthcare, climate change, education and economic development.
• A Rust-and-Arrow approach for Apache Spark reportedly runs twice as fast as a similar PyArrow source and nearly 20 times faster than native Python without Arrow.
• Rohan Pandey’s Bayesian-optimization agent tested 50 GPT OSS 120B serving configurations and reported a 17.5% goodput-token improvement.
• Cybersecurity expert Brett Goldstein warns that AI can expose weaknesses in outdated software and urges organizations to use AI to identify and fix vulnerabilities while strengthening security practices.

Preparing for AI agent incidents requires more than containment

AI agent incidents can involve mistakes rather than an outside attacker, so standard security playbooks may send teams looking for someone who was never there. Organizations should decide in advance how to classify incidents by their impact, who can make that call, and which events start reporting deadlines. Agents can take many actions quickly, so teams need records that show which machine identity acted and how agents handed work to each other. They should also track what each tool can change and whether those changes can be undone.

AI agent offboarding means revoking access and preserving its work

Deleting an AI agent from a dashboard does not remove its credentials or explain the work and decisions it left behind. Agents may use shared tokens, hold service accounts, schedule future jobs, or pass tasks to other agents, so teams should trace access and review unfinished work before shutdown. Short-lived credentials limited to specific actions can reduce leftover access. Organizations should also preserve the model and prompt versions, data, tools, and policies behind past decisions, since those decisions may be questioned years later.

UK GDP grew 0.7% as computer related activity added to growth

UK real GDP grew 0.7% in the third quarter of 2026, following 0.5% growth in the second quarter, according to the Office for National Statistics estimate published on 12 November. Services output rose 0.9%, production grew 0.1%, and construction fell 0.7%. Manufacturing grew 0.9%, with computer, electronic and optical products up 4.6%; computer programming and consultancy added 0.15 percentage points to GDP growth. Business investment increased 0.4%, below the Bank of England’s projection of around a further 1%. The ONS did not separate AI related growth from other activity.

Google introduces Gemini smartglasses starting at $395

Google is preparing to launch Gemini powered smartglasses, with prices starting at $395. The product faces concerns about comfort and privacy, issues that have affected smartglasses since the Google Glass era. Google is emphasizing privacy as it makes another attempt to gain acceptance for the devices.

Cornell plans five summits on AI and data science

Cornell University will hold five Data Science and Artificial Intelligence Thought Summits to explore new questions and challenges and encourage collaboration among experts. The topics are social good, healthcare, climate change, education, and economic development. The events will include keynote speakers, panel discussions, breakout sessions, and networking opportunities. Cornell says the summits aim to support interdisciplinary research and develop solutions in data science and AI.

Rust and Arrow help Spark skip a costly data detour

A new approach pairs a Rust decoder with a thin Python layer to send data into Apache Spark without re-encoding it or creating Python objects for each row. It uses PyO3, Maturin, arrow-rs and the Arrow C Data Interface to build Arrow RecordBatches that Spark can send directly to Delta tables. Benchmarks cited in the post found it ran twice as fast as a similar pure Python source using PyArrow and nearly 20 times as fast as native Python without Arrow. The Spark API shipped in Apache Spark 4.0 and is generally available from Runtime 15.4 LTS and Serverless environment 2; the post also points to an ARINC 429 example repository.

Rohan Pandey’s agent improves VLM serving settings

Rohan Pandey demonstrated an agent that uses Bayesian optimization to tune settings for serving vision language models, replacing manual trial and error. In a GPT OSS 120B test on a droplet, tuner.py tested 50 configurations across chat short, chat long, rag and batch classify over about four to five hours. Pandey reported that the best setup increased goodput tokens by 17.5 percent from the starting configuration. The workflow returns an HTML report and an optimized YAML file, and the article says DigitalOcean is also building benchmarking and regression tools for TTFT, ITL, throughput and accuracy checks.

AI tools expose deep security weaknesses across the internet

Brett Goldstein, a Vanderbilt University professor who specializes in cybersecurity and artificial intelligence and a former Pentagon official, warns that AI tools are exposing weaknesses caused by outdated software practices. He says hackers could exploit these flaws and cause a broad failure of internet systems. Goldstein calls for using AI tools to find and fix vulnerabilities, along with new security protocols and practices. He describes internet security as critical infrastructure and a matter of national security.

Local businesses and students weigh AI’s role in branding

The Daily Nebraskan spoke with local business owners and students about using AI in branding and marketing, where opinions differ. Emily Wilson, owner of Emily’s Boutique, said AI helps with tasks such as social media management but should not replace human creativity or a unique, authentic brand. John Lee, owner of JL Marketing, said AI can save time and money on data analysis and content creation. He also said it can help identify consumer trends and patterns to guide marketing plans.

Survey finds Michigan small businesses use AI and see risks

A Michigan Small Business and Technology Coalition survey found that 71% of respondents use AI in some way, and 71% believe it will have a positive impact on their business. Among the 150 small business owners and entrepreneurs surveyed, 45% use AI to improve customer service and 31% use it to develop products. At the same time, 62% see AI as a risk, with 45% concerned about data security and 31% concerned about job displacement. The survey was conducted in October and November, and its results were released on November 12.

Sources

• AI Agent Incident Response
• Who collects the keys when an AI agent retires?
• UK Grows 0.7% on AI-Linked IT; Business Investment Lags BoE
• Google pushes privacy angle as Gemini-powered smartglasses debut starting at $395
• Five Cornell Thought Summits to explore emerging questions in AI and data science
• Rust plus Arrow beats the Lambda detour
• Rohan Pandey built an agent that tunes VLM serving configs
• How the internet could break
• To prompt or not to prompt; local businesses, students talk AI branding
• Michigan small businesses embrace AI