AI News of January 14, 2026: Gemini Personal Intelligence, GPT-5.2-Codex API

This week in AI

Busy day for the AI ecosystem: Google unveils Personal Intelligence in Gemini, enabling secure connection to Google apps for truly personalized answers. OpenAI makes GPT-5.2-Codex accessible via its Responses API, while NotebookLM rolls out Data Tables for all users.

Google launches Personal Intelligence in Gemini

January 14, 2026 — Google announces Personal Intelligence, a new feature that allows Gemini to securely connect to user’s Google apps for truly personalized answers.

🔗 X Post @GeminiApp | Google Blog

Connectable applications

Gemini can now pull insights from:

Application	Accessible Data
Gmail	Emails, conversations
Google Photos	Images, memories
Google Search	Search history
YouTube	History, preferences

Use cases

Google highlights two concrete examples:

Planning: Gemini suggests destinations and “hidden gems” adapted to your preferences for your professional or personal trips
Shopping: Gemini learns your tastes and preferences to help you find items that really match you

Privacy first

Google insists on user control:

OFF by default: the feature must be manually activated
Granular control: you choose exactly which apps to connect
Reversible: you can disable it at any time

Availability

Aspect	Details
Status	Beta
Region	United States only
Subscriptions	Google AI Pro and Ultra
Expansion	Planned as the feature improves

Personal Intelligence begins rolling out today as a beta to Google AI Pro and Ultra subscribers in the U.S. As the feature improves, availability will expand. — @GeminiApp

GPT-5.2-Codex available in Responses API

January 14, 2026 — OpenAI announces the availability of GPT-5.2-Codex in its Responses API, the same model used in Codex.

🔗 X Post @OpenAIDevs | Documentation

Capabilities

GPT-5.2-Codex excels in complex and long-running development tasks:

Task	Description
Building features	Construction of complete new features
Refactoring	Restructuring and improving existing code
Bug hunting	Identification and correction of bugs
Security	Detection of vulnerabilities in codebases

The most cyber-capable

OpenAI highlights that GPT-5.2-Codex is their most capable model for cybersecurity, able to:

Find vulnerabilities in codebases
Understand security flaws
Suggest corrections

Context

GPT-5.2-Codex was launched on December 18, 2025, exclusively in Codex. This API opening now allows developers to integrate this model into their own applications and workflows.

NotebookLM rolls out Data Tables to everyone

January 14, 2026 — NotebookLM announces the general rollout of Data Tables, a feature that automatically transforms notes into structured tables.

🔗 X Post @NotebookLM

Prompt examples

NotebookLM proposes templates according to use cases:

Domain	Suggested Prompt
Work	Convert my meeting notes into a table with columns: action, category, priority, owner
Science	Create a table of clinical trial results: Study, Year, Method, Sample size, Statistics
Travel	Table of destinations: City, Best time, Estimated cost/day
School	Table of historical events: Name, Country, Date, Key figures, Economic consequences

What changes

Data Tables allows moving instantly from raw notes to structured and actionable data, ideal for analysis and comparison.

Anthropic supports the ARPA-H PCX program

January 14, 2026 — Anthropic announces its support for the ARPA-H PCX (Pediatric Care eXpansion) program, a $50 million effort to improve pediatric care.

🔗 X Post @AnthropicAI | ARPA-H Release

The PCX program

Aspect	Details
Budget	$50 million
Scope	200+ pediatric hospitals
Initial focus	Pediatric brain cancer
Goal	Reduce the care journey from years to weeks

How it works

The program aims to share data between hospitals on complex cases, allowing doctors to learn from similar cases treated elsewhere.

We’re supporting ARPA-H’s PCX program—a $50M effort to share data between 200+ pediatric hospitals on complex cases, beginning with pediatric cancer. The goal is to help doctors learn from similar cases and shorten the care journey from years to weeks. — @AnthropicAI

ElevenLabs: 230 customer interviews in 24h with Agents

January 13, 2026 — ElevenLabs shares an impressive use case of its conversational Agents: 230 customer interviews conducted in less than 24 hours for their app Eleven Reader.

🔗 X Post @elevenlabsio | Agents Documentation

The problem solved

Approach	Advantages	Disadvantages
Live interviews	Deep insights	Doesn’t scale
Surveys	Scales easily	Loses nuance
Conversational Agents	Depth + Scale	Best of both worlds

The quantified results

Metric	Result
Completed interviews	230
Total time	< 24 hours
Success rate	85% on-topic and successful
Average duration	10 minutes per call
Time-to-production	Insights delivered the next day

Rapid deployment

The interviewer agent was built and deployed in less than a day. ElevenLabs used the Analysis feature to evaluate each call and extract structured data from transcripts.

Key learnings

95% of respondents interacted naturally without recognizing they were speaking to an AI
21% of fiction readers asked for multi-character support for dialogues
One user: “This customer service interview is the most remarkable AI experience I have ever had”

What it demonstrates

Conversational AI agents allow conducting user research on a global scale, 24/7, while maintaining the richness of qualitative exchanges.

What it means

Google’s Personal Intelligence represents a major evolution of AI assistants. By connecting Gemini to the user’s personal data, Google creates a truly contextual assistant. Privacy will be the key adoption factor.

The GPT-5.2-Codex API democratizes access to OpenAI’s best coding model. Developers can now build AI-augmented development tools without going through Codex. The focus on cybersecurity is notable.

NotebookLM’s Data Tables confirms Google’s strategy to transform NotebookLM into a complete productivity tool, not just an AI podcast generator.

Anthropic’s engagement in healthcare via ARPA-H shows a different approach from OpenAI: institutional partnerships vs startup acquisitions.

ElevenLabs demonstrates with concrete metrics the maturity of its conversational agents: 85% success rate, deployment in less than a day, and above all 95% of users who did not recognize speaking to an AI. It is a reproducible use case for any company looking to scale its user research.