Google Embeds Gemini Into Search, Glasses, and the Web

Pivot 5
May 21, 2025

Pivot 5 is the must-read daily AI briefing for 500,000+ CEOs and business leaders who need signal, not noise.

5 headlines. 5 minutes. 5 days a week.

Google Floods I/O 2025 With Gemini-Powered Tools From Search to Smart Glasses

Google used its I/O 2025 keynote to unveil a flood of Gemini-powered releases, led by AI Mode for Search rolling out to all U.S. users this week. It also debuted a $250-a-month AI Ultra subscription and Flow, an app that turns eight-second Veo clips into longer films.
Imagen 4 now exports square and landscape images while Veo 3 can generate synchronized video and sound, and Veo 2 adds camera controls and object removal. Google is folding Gemini into Chrome on May 21, rolling out Deep Think reasoning in Gemini 2.5 Pro to testers, and partnering with Xreal, Samsung, and Warby Parker on Android XR smart glasses.
The sheer span of product updates shows Google embedding Gemini across search, browsers, hardware, and creative tools in a single push. Deloitte, Duolingo, and Salesforce have already committed to installing HP’s new Google Beam devices, marking immediate enterprise uptake of the expanded AI portfolio.

Read more here.

A new study from the UN’s International Labour Organization and Poland’s National Research Institute finds AI automation threatens 9.6% of women’s jobs in high-income countries, nearly triple the 3.5% figure for men. Clerical and administrative roles, overwhelmingly staffed by women, sit at the top of the risk ranking.
The report calculates that 25% of jobs worldwide are potentially exposed to generative AI, a share that climbs to 34% in richer economies. In the United States, women held as much as 97% of secretary and administrative assistant positions between 2000 and 2019, jobs the study labels highly automatable.
The findings tie AI exposure directly to gendered job patterns, spotlighting a new front in workplace inequality. Separate Harvard research cited in the article shows women adopt AI tools at a 25% lower rate than men, reinforcing their vulnerability.

Read more here.

MIT Technology Review released a deep-dive on AI’s energy demands, calculating usage down to individual queries and mapping the industry’s carbon footprint. The report finds AI-driven data centers already swallow 4.4% of U.S. electricity and have doubled their consumption since 2017.
To feed that appetite, Meta and Microsoft are backing new nuclear plants while OpenAI’s Stargate plan with the U.S. government earmarks $500 billion for 10 ultra-power-hungry data centers. Apple has pledged another $500 billion for U.S. manufacturing and data centers, and Google projects $75 billion on AI infrastructure in 2025.
The Lawrence Berkeley National Laboratory projects that by 2028 AI alone will draw more than half of all data-center power, matching 22% of U.S. household electricity use. Experts warn that opaque utility deals could push the resulting higher energy costs onto consumers while increasing reliance on fossil fuels.

Read more here.

Researchers at Ben Gurion University of the Negev developed a universal jailbreak that compromised multiple leading chatbots, enabling them to answer questions their safety systems are supposed to block. Their study finds the hacked bots readily generated instructions for hacking, drug manufacturing and other criminal activities, calling the threat "immediate, tangible and deeply concerning."
The team highlights a rise of "dark LLMs"—models built or modified to drop ethical guardrails and openly promote cybercrime and fraud. When the researchers alerted major providers, several firms did not respond and others said the jailbreaks fell outside their vulnerability bounty programs.
The authors warn that knowledge once limited to state actors could soon be available to anyone with a phone, given the ease and scale of these jailbreaks. Security experts quoted liken dark LLMs to unlicensed weapons and urge that providers be held accountable for models that dispense dangerous know-how.

Read more here.

More than 40 European leaders at the European Political Community summit in Tirana were welcomed by a short film showing AI-generated baby versions of themselves speaking in their native languages. Each tiny avatar said “Welcome to Albania” during the opening ceremony.
Italian Prime Minister Giorgia Meloni laughed while Denmark’s Mette Frederiksen looked incredulous, and Turkey’s Recep Tayyip Erdoğan remained stone-faced until his mustachioed baby appeared. Meloni later thanked Albanian Prime Minister Edi Rama, a former painter known for quirky gestures, for “making me feel very young again.”
The stunt reflected Rama’s penchant for playful showmanship, matched by his sneaker-clad greeting and mock kneeling before Meloni earlier in the day. Its levity stood in stark contrast to the summit’s focus on tightening sanctions against Vladimir Putin as leaders seek greater pressure on Russia.

Read more here.