On Friday, December 5, 2025, artificial intelligence news is dominated by three big themes: more powerful reasoning models, billion‑dollar “agentic AI” bets in the enterprise, and mounting concern over AI security and governance.
From Google’s new Gemini 3 Deep Think mode to Anthropic’s $200 million “agentic AI” deal with Snowflake, and new guidance from security agencies on how not to use AI in critical infrastructure, today’s headlines show an industry that’s both accelerating and learning to hit the brakes.
Below is a roundup of the most notable AI stories published or updated on December 5, 2025, curated and explained for readers of Google News and Discover.
1. Google’s Gemini 3 Deep Think aims to win the reasoning race
Google has rolled out “Gemini 3 Deep Think,” described as its most advanced AI reasoning mode so far, inside the Gemini app for Google AI Ultra subscribers. [1]
According to India Today, Deep Think is designed specifically for multi‑step, analytical problems in areas like mathematics, logic, and science. Instead of taking a single linear path through a question, the mode can explore multiple lines of reasoning in parallel before deciding on the most plausible answer. [2]
Key points from Google’s rollout:
- Deep Think is a mode on top of Gemini 3, not a separate model. Users enable it from the prompt bar when using Gemini 3 Pro. [3]
- Google reports strong results on reasoning benchmarks such as “Humanity’s Last Exam” and ARC‑AGI‑2, calling the scores an “unprecedented milestone” in tool‑free logical reasoning. [4]
- The feature is explicitly targeted at researchers, professionals, and students who need detailed, step‑by‑step explanations, not just quick answers. [5]
This release arrives just as Google’s wider Gemini 3 family has been getting unusually glowing reviews. AI pioneer Geoffrey Hinton told Business Insider that after Gemini 3 and Google’s Nano Banana Pro image model, Google is “beginning to overtake” OpenAI, partly because it designs its own AI chips and runs massive data centers in‑house. [6]
Combined with earlier reports that OpenAI CEO Sam Altman has internally raised a “code red” over Gemini 3’s momentum, the Deep Think launch underscores how reasoning performance has become the new front line in the Google–OpenAI rivalry. [7]
2. OpenAI focuses on infrastructure, travel, and a smarter AI browser
While Google pushes on reasoning benchmarks, OpenAI’s news today is more about infrastructure and user experience.
2.1. NEXTDC + OpenAI: a hyperscale AI campus in Sydney
In Australia, data‑center operator NEXTDC announced a memorandum of understanding with OpenAI to build a hyperscale AI campus and GPU supercluster at its S7 site in Western Sydney, with potential capacity of up to 550 megawatts. [8]
The Reuters report notes:
- NEXTDC shares jumped more than 10% on the announcement, making it the top gainer on Australia’s benchmark index despite a broader market decline. [9]
- The campus is part of OpenAI’s “Australia program,” pitched as a blueprint to boost AI infrastructure and adoption across the country. [10]
- Sydney recently shifted from a more restrictive AI stance to a roadmap emphasizing economic growth through AI adoption, making the deal politically and economically aligned with local priorities. [11]
The project highlights how GPU campuses and energy‑hungry AI clusters are becoming central infrastructure bets, not just for hyperscalers but for regional data‑center specialists.
2.2. ChatGPT Atlas AI browser gets a major upgrade
On the consumer side, OpenAI has quietly upgraded ChatGPT Atlas AI, its AI‑powered Mac browser launched last month. According to AffiliateBooster, the latest update focuses heavily on usability and deeper integration between browsing and ChatGPT. [12]
The new version introduces:
- Vertical tabs and multi‑tab selection for easier navigation.
- Extension importing, allowing users to bring Chrome or Edge extensions into Atlas AI with minimal friction.
- Google as a default search engine option instead of being locked to Bing.
- A more powerful “Ask ChatGPT” sidebar that can now insert AI responses directly into forms and emails with a single click.
- An “Agent mode” (preview for Plus/Pro) so ChatGPT can handle tasks like form‑filling, research aggregation, or travel booking across multiple tabs once the user approves. [13]
For Mac users, this positions Atlas AI as a full replacement for traditional browsers, not just a wrapper around ChatGPT.
2.3. AI‑powered travel with Virgin Australia
In travel, Travel and Tour World reports that Virgin Australia is teaming up with OpenAI to launch AI‑powered flight search and personalised trip‑planning tools, aiming to make booking and itinerary planning more conversational and proactive for travellers. [14]
Together with infrastructure deals like NEXTDC and product tweaks like ChatGPT Atlas AI, today’s announcements reinforce OpenAI’s strategy: embed its models deeply into both the cloud stack and everyday consumer experiences.
3. Anthropic doubles down on “agentic AI” and user research
If OpenAI is building the highways, Anthropic is trying to own the vehicles that drive on them.
3.1. A $200 million “agentic AI” deal with Snowflake
IT Pro reports that data‑cloud giant Snowflake has signed a multi‑year, $200 million expansion of its partnership with Anthropic to bring what it calls “agentic AI” to more than 12,000 enterprise customers. [15]
Highlights from the deal:
- Anthropic’s Claude Sonnet 4.5 and Claude Opus 4.5 models will be tightly integrated into Snowflake Cortex AI to power “Snowflake Intelligence”, an enterprise AI agent that sits directly on top of customers’ governed data. [16]
- The approach keeps sensitive data inside Snowflake’s environment, addressing CIO fears about exporting information to external AI services. [17]
- Existing joint customers already process trillions of Claude tokens per month via Snowflake, according to the article, and the new deal is meant to move more use cases from pilot to production. [18]
Anthropic CEO Dario Amodei framed the partnership as “bringing AI to where the data lives,” a core theme in enterprise AI in 2025.
3.2. Anthropic Interviewer: a bot that interviews you about AI
On the research side, Anthropic today launched “Interviewer”, a conversational bot designed to quiz users about how they use AI and how they feel about it. [19]
According to Dataconomy:
- Interviewer runs as a short‑term pilot inside Claude, popping up to ask users about their workflows, expectations, and concerns. [20]
- The system uses Claude to draft questions, conduct interviews, and help analyze the transcripts, with human researchers still supervising and interpreting the results. [21]
- In a preliminary study of 1,250 professionals, 86% said AI saves them time, and about two‑thirds described their current AI use as augmentation (AI assisting humans) rather than full automation. [22]
The tool underscores a growing realization among AI labs: winning the race isn’t only about model capabilities—it’s also about understanding human comfort, trust, and expectations.
4. Is there an AI bubble? Markets send mixed signals
Today’s financial headlines show investors wrestling with a big question: are we in an AI bubble—or just the first act of a much longer boom?
4.1. NTT DATA: a “short‑lived bubble” with a stronger rebound
In an interview with Reuters, NTT DATA CEO Abhijit Dubey argued that if there is an AI bubble, it will “deflate faster” than previous tech crazes but lead to an even stronger rebound as real corporate use catches up with infrastructure spending. [23]
Dubey expects:
- A 12‑month “normalisation” period in valuations.
- Persistent constraints on compute supply, giving chipmakers and hyperscalers significant pricing power.
- Long‑term, AI as a “massive secular trend” comparable to the rise of the internet. [24]
He also flagged a growing skills shift, with companies rethinking hiring strategies as AI changes the demand for different types of labor. [25]
4.2. Asia’s equity pipeline and AI‑driven IPO risk
A second Reuters piece looks at Asia’s booming equity capital markets, where IPO and follow‑on deals have climbed to about $267 billion in 2025, the first annual increase since 2021. [26]
At the same time:
- A wave of AI‑centric companies—Chinese LLM builders like Zhipu AI and MiniMax, and AI‑chip makers like MetaX and Kunlunxin—is preparing multi‑billion‑dollar listings. [27]
- Bankers warn that if AI valuations correct sharply, deal pricing could suffer across the board, with contagion risk beyond the tech sector. [28]
The takeaway: capital is still flooding into AI, but even bullish voices admit that 2026 could be bumpy if sentiment swings.
4.3. Son & Hinton: super‑AI optimism and a bet on Google
In parallel, two of AI’s most influential figures weighed in:
- Masayoshi Son, CEO of SoftBank, said at an event today that future “super AI” systems could surpass humans to such an extent that people seem “like fish,” even suggesting such systems could one day win a Nobel Prize for discoveries beyond human reach. [29]
- Geoffrey Hinton, often called the “Godfather of AI,” told Business Insider he now expects Google to beat OpenAI in the AI race, praising Gemini 3 and Google’s custom chip strategy as key advantages. [30]
Together, these comments capture the mood of late 2025: concern about frothy valuations, but near‑religious belief among some leaders that AI’s long‑term upside is enormous.
5. Global adoption and geopolitics: India surges, North Korea weaponizes AI, and the Vatican calls for responsibility
AI adoption is not happening evenly—and today’s stories underline those fault lines.
5.1. India leads the world in generative AI usage
A new OECD–Cisco study covered by The Indian Express finds that India now leads all surveyed countries in active generative AI use, with:
- 66.4% of respondents in India using AI tools actively.
- Over 89% saying AI is useful, and more than 84% saying they trust it.
- Around 78% reporting they’ve taken some AI training, and more than half planning further training in the next year. [31]
By contrast, people over 45 in many countries report far lower adoption and weaker trust, suggesting a widening generational and geographic gap in AI familiarity. [32]
The same study notes that users in India, Brazil, Mexico and South Africa also report higher recreational screen time, raising fresh questions about digital well‑being alongside AI skills. [33]
5.2. North Korea’s use of generative AI for hacking and research
At the other end of the spectrum, a report in the Korea JoongAng Daily says North Korea is using generative AI tools such as ChatGPT to train professional hackers and conduct scientific research at universities. [34]
Researchers at a South–North broadcasting and telecoms conference concluded that:
- ChatGPT‑style systems are likely being used under tight surveillance to train elite IT workers and infer what kinds of data South Korea values. [35]
- North Korea’s domestic smartphones (like recent Jindallae models) now have AI capabilities as well as enhanced surveillance features, including automatic screenshots and watermarking to trace file sharing. [36]
It’s a stark reminder that the same AI tools enabling productivity in one country can be turned into cyber‑weapons in another.
5.3. Pope Leo XIV: “New generations must be helped, not hindered”
In Rome, Pope Leo XIV addressed participants at the conference “Artificial Intelligence and Care for Our Common Home,” warning that young people need guidance, not obstacles, in their relationship with AI and new technologies. [37]
Key themes from his remarks, reported by Vatican News:
- Access to vast data is not the same as wisdom or meaning; society must help youth learn to interpret information and ask deeper questions. [38]
- The Pope called for coordinated action from politics, business, finance, education, media, and religious communities to ensure AI serves the common good rather than concentrating wealth and power “in the hands of a few.” [39]
- He emphasized that humans should be “co‑workers in the work of creation,” not passive consumers of AI‑generated content, and urged special attention to the neurological and emotional development of children in an AI‑saturated environment. [40]
Ethical debates around AI are clearly no longer confined to tech conferences—they’re now a fixture in global religious and civic forums.
6. Security: new guidance for critical systems and a real‑world prompt‑injection attack
With AI creeping into everything from industrial control systems to developer tools, security stories are becoming a daily part of AI news.
6.1. NSA warns: don’t let LLMs make safety‑critical decisions
CSO Online reports that the US National Security Agency (NSA) and partner agencies in Australia and elsewhere have published “Principles for the Secure Integration of Artificial Intelligence in Operational Technology.” [41]
The guidance focuses on operational technology (OT)—the systems that control energy plants, water treatment facilities, hospitals, and factories—and highlights:
- Risks from prompt injection, data poisoning, and “AI drift” (where models become less accurate as conditions change). [42]
- The danger of over‑reliance on AI, which can deskill human operators and overload them with alerts. [43]
- A clear warning that LLMs and similar AI systems are “almost certainly” not reliable enough to make safety‑critical decisions independently in OT environments. [44]
In short: AI may help monitor and optimize critical infrastructure, but humans must remain firmly in charge of the off switch.
6.2. PromptPwnd: Fortune 500 CI/CD pipelines exploited via prompt injection
On the software side, Cyber Press details a newly publicized vulnerability pattern dubbed “PromptPwnd.” [45]
Security firm Aikido found that some GitHub Actions and GitLab CI/CD pipelines:
- Take untrusted user content (like issue titles or pull‑request descriptions).
- Feed that content directly into AI prompts used by automation agents (e.g., Gemini CLI, Claude, OpenAI‑based tools).
- Grant those agents high‑privilege tokens, including repository write access and cloud credentials. [46]
Attackers can hide malicious instructions in seemingly normal text, causing the AI agent to:
- Execute arbitrary commands.
- Leak secrets such as API keys, access tokens, or GitHub tokens. [47]
At least five Fortune 500 companies were affected, and even Google’s own Gemini CLI repo had to be patched after Aikido demonstrated the issue. [48]
Recommended mitigations include:
- Sanitizing all user input before sending it to AI agents.
- Restricting what tools agents can call.
- Treating AI automation like any other code path, with rigorous security review. [49]
Combined with the NSA guidance, today’s stories send a clear message: AI security is no longer hypothetical—real exploits are here.
7. AI quietly embeds itself into customer service and gaming
Not all AI news today is about geopolitics and billion‑dollar data centers—some of it is about the mundane systems that keep digital businesses running.
One example: iGaming supplier EveryMatrix has announced it is integrating AI agents into its player account management and back‑office systems. [50]
According to InterGame:
- The AI agents, built with partner Cevro AI, will handle tasks like withdrawals, deposit issues, bonus requests, and account questions 24/7. [51]
- The company stresses that AI behavior will be fully aligned with responsible gambling and GDPR requirements, highlighting how compliance is now a sales point for AI vendors, not just a constraint. [52]
Paired with travel‑planning bots, AI‑assisted medical workflows (such as recent work on anesthesiology decision support), and AI‑enhanced browsers, these developments show how AI is dissolving into the background of everyday digital services, even when it’s not in the headline.
8. What today’s AI news means for you
Putting all of these stories together, a few patterns stand out:
- Reasoning and “agentic AI” are the new battlegrounds.
Google’s Deep Think mode, Anthropic’s Snowflake partnership, and OpenAI’s browser‑integrated agents all point toward AI systems that reason over long‑context tasks and act on your behalf, not just chat. [53] - Infrastructure and geography matter more than ever.
Deals like NEXTDC–OpenAI in Sydney and sovereign‑cloud pushes in Europe illustrate that physical data centers, energy supply, and local regulations are shaping where AI power actually lives. [54] - Adoption is skyrocketing—but unevenly.
India and other emerging economies are racing ahead in everyday AI use, while older and wealthier populations move more slowly and often more skeptically. That creates both opportunities and new digital divides. [55] - Security and governance are catching up, fast.
From the NSA’s OT guidelines to PromptPwnd in CI/CD pipelines, today’s stories show that AI security is now a mainstream concern, not a niche topic for researchers. [56] - Ethics are moving into the spotlight.
The Vatican’s call to “help, not hinder” younger generations, and public debates about North Korea’s use of AI for hacking, underscore that questions of control, fairness, and human dignity are becoming just as important as benchmarks and funding rounds. [57]
For businesses, the implication is simple: AI is now a board‑level topic that touches strategy, infrastructure, compliance, security, and culture all at once. For individuals, today’s news is a reminder to build AI literacy—not only to keep up, but to understand where and how these systems are shaping your work, your data, and your society.
References
1. www.indiatoday.in, 2. www.indiatoday.in, 3. www.indiatoday.in, 4. www.indiatoday.in, 5. www.indiatoday.in, 6. www.businessinsider.com, 7. fortune.com, 8. www.reuters.com, 9. www.reuters.com, 10. www.reuters.com, 11. www.reuters.com, 12. www.affiliatebooster.com, 13. www.affiliatebooster.com, 14. www.travelandtourworld.com, 15. www.itpro.com, 16. www.itpro.com, 17. www.itpro.com, 18. www.itpro.com, 19. dataconomy.com, 20. dataconomy.com, 21. dataconomy.com, 22. dataconomy.com, 23. www.reuters.com, 24. www.reuters.com, 25. www.reuters.com, 26. www.reuters.com, 27. www.reuters.com, 28. www.reuters.com, 29. www.digitaljournal.com, 30. www.businessinsider.com, 31. indianexpress.com, 32. indianexpress.com, 33. indianexpress.com, 34. koreajoongangdaily.joins.com, 35. koreajoongangdaily.joins.com, 36. koreajoongangdaily.joins.com, 37. www.vaticannews.va, 38. www.vaticannews.va, 39. www.vaticannews.va, 40. www.vaticannews.va, 41. www.csoonline.com, 42. www.csoonline.com, 43. www.csoonline.com, 44. www.csoonline.com, 45. cyberpress.org, 46. cyberpress.org, 47. cyberpress.org, 48. cyberpress.org, 49. cyberpress.org, 50. www.intergameonline.com, 51. www.intergameonline.com, 52. www.intergameonline.com, 53. www.indiatoday.in, 54. www.reuters.com, 55. indianexpress.com, 56. www.csoonline.com, 57. www.vaticannews.va
