OpenAI Ships GPT-6 Cyber as Google Extends Gemini Omni and Claude Opens Personal Finance

OpenAI Introduces GPT-6 Cyber for Scoped Vulnerability Research

OpenAI on Thursday introduced GPT-6 Cyber, a cybersecurity model trained to find and validate exploitable vulnerabilities while staying inside researcher-set scope. OpenAI’s GPT-6 Cyber announcement reports a 48.5% score on ExploitGym, up from 32.1% for GPT-5.6 Cyber, a roughly 51% relative improvement. On a separate alignment evaluation, the model avoided exploiting out-of-scope honeypot services in 99.93% of tests without its deployment prompt and 100% with that prompt. OpenAI classifies the model as Critical in cybersecurity under its Preparedness Framework, up from High for GPT-5.6 Cyber.

On a new web penetration-testing recall evaluation, GPT-6 Cyber scored 40.3% versus 32.0% for GPT-6 Astra. On RemoteExploitBench, Astra with Daybreak Blue refused 100% of tasks while Cyber completed 95%. The company says the model helped identify 81 previously unknown high-severity Firefox vulnerabilities, with Mozilla shipping fixes for many of them, and that the same model powered the mid-October Linux local privilege-escalation research we covered yesterday. GPT-6 Cyber starts with select Daybreak Red API and Codex customers, with broader Daybreak Red and Blue expansion planned; OpenAI is also hosting a Cyber at the Frontier livestream on November 12 at 10 a.m. PT. Firefox Distinguished Engineer Brian Grinstead put the customer case this way:

“GPT‑6 Cyber is a world-class model for cyberdefense. Firefox entered 2026 as one of the world’s most hardened and scrutinized codebases. Even after further hardening with multiple generations of frontier models, GPT‑6 Cyber uncovered previously unknown high-severity vulnerabilities, identified architectural improvements, and helped us ship fixes to millions of users.”

Google Ships Gemini Omni 1.2 Flash With 15-Second Video

Google DeepMind rolled out Gemini Omni 1.2 Flash and made it generally available on the Gemini API and Google Cloud. Google’s Gemini Omni 1.2 Flash announcement says native video length rises to 15 seconds from 10, with duration selectable in one-second increments from 3 to 15 seconds. API pricing drops 30% from Omni 1.1 to $0.07 per second at 720p. The release adds multi-turn conversational editing so creators can swap products, backgrounds, lighting or aspect ratio without regenerating from scratch, plus stronger reference adherence for characters, products and style images, including short motion clips up to three seconds and audio cues.

Google says drafts can start at 360p or 720p, then upscale winners to 1080p or 4K, and that outputs carry an imperceptible SynthID watermark. Availability spans Google AI Studio and the Gemini API for developers, Google Cloud for business, Flow for scene direction, Google Vids for 15-second clips at no additional cost with a Google account or Workspace subscription, the Gemini app for Google AI Ultra subscribers as it rolls out, and Google Ads for campaign scale. The company notes ongoing work on fast sports action, rapid script changes and complex physics.

Anthropic Opens Personal Finance in Claude for US Subscribers

Anthropic began rolling out Personal Finance in Claude as a beta this week for Pro, Max and Ultra plans in the United States. Anthropic’s Personal Finance in Claude announcement says users can securely connect bank and card accounts through Plaid with read-only access. Claude cannot move, manage or invest money, and Personal Finance has no ads or sponsored product recommendations. Anthropic says conversations inside the Personal Finance hub are not used to train models, users approve access without sharing bank passwords with Claude, and accounts can be unlinked at any time.

For connected accounts, Claude can see the bank name, account name, type, last four digits, balances and credit limits, and up to two years of transactions. The product pitches natural-language questions about spending, upcoming payments and simple dashboards, while warning that Claude can make mistakes and that important figures should be checked with the bank. Anthropic frames the launch as a first step toward working alongside licensed professionals for larger money decisions, and says it wants to hear from banks and wealth managers. Reuters’ report on Claude Personal Finance confirms the US beta rollout for paid plans.

METR Finds Roughly 2× AI R&D Acceleration at Anthropic

METR published an independent assessment of AI R&D acceleration at Anthropic after a high-access evaluation from August 12 to October 29, 2026. METR’s Anthropic AI R&D acceleration assessment says most staff on the project now believe Anthropic is improving capabilities on well-specified tasks at around twice the rate seen between March 2024 and October 2025, with a plausible range nearer 1.3× to 3×. Since October 2025, Anthropic’s AECI benchmark trend has grown at an estimated 1.9× the prior rate with a 95% confidence interval of 1.5× to 2.8×. METR is confident the acceleration was not due to faster training-compute scaling.

The assessors guess that a majority—perhaps about 80%—of the current October 2026 acceleration is attributable to AI use in R&D, with substantial uncertainty and other factors more important in late 2025. They did not identify clear bottlenecks over the next 12 months that would prevent further compounding, while noting their understanding of bottlenecks changed several times and that Anthropic redacted a majority of the evidence METR found most informative. A companion review of the AI R&D section of Anthropic’s August 2026 Risk Report was published separately.

Mistral Expands Studio With Managed Agents, Apps and Workflows GA

Mistral expanded Mistral Studio alongside Mistral Large 4, also called Le Chonk. Mistral’s Studio expansion announcement says managed agents and apps enter public preview while workflows reach general availability, with new enterprise context, agent observability and evaluations. Studio can deploy in cloud, VPC or on-premises environments, connect to enterprise systems through connectors and MCP servers, and host agents with skills, sandboxes, subagents and approval hooks. Customers and partners named in the post include ABANCA, Accenture, Alten, BNP Paribas, Capgemini, CMA CGM and Tata Consultancy Services.

Mistral describes internal examples such as an incident-investigation agent that follows runbooks and posts findings to Slack, a product-operations workflow that routes bugs and feature requests, and a staffing app that recommends people with supporting evidence. OpenTelemetry traces and evaluation datasets are positioned as operator-owned knowledge. The company pitches sovereignty: choose where models and the platform run, govern tool access, and require human approval for sensitive actions.

OpenAI and JPMorganChase Announce a Strategic Partnership

OpenAI president Greg Brockman announced a strategic partnership with JPMorganChase. In Greg Brockman’s X post announcing the JPMorganChase partnership, he said the companies will explore how advanced AI can support the bank’s enterprise by helping engineers build, security teams defend, and employees and customers benefit from more capable AI, with a particular focus on cybersecurity for the financial system. Brockman made related remarks at the WSJ Tech conference around November 4. PYMNTS’ secondary report on the OpenAI–JPMorgan partnership, citing The Wall Street Journal, says the bank will get access to cybersecurity-focused models through OpenAI’s Daybreak program; we are not inventing deal value, seat counts or delivery timelines beyond what those sources state. Brockman wrote:

“Announcing our strategic partnership with JPMorganChase, to explore how advanced AI can support their enterprise: helping engineers build, security teams defend, and employees and customers benefit from more capable AI. I’m especially excited about working together on cybersecurity, to help secure and protect the financial system to a level that is only possible with AI.”

What We Left Out

AI at Meta’s X post on Muse Realtime Avatar says Meta is rolling Muse Realtime Avatar out in the Muse app in the US after previewing it at Connect, pointing readers to the research write-up. That research post, Meta’s Muse Realtime Avatar research blog, is a September 23, 2026 technology piece on audio-driven Diffusion Transformer avatars, subsecond latency and Meta Video Seal watermarking; we are not inventing a broader product rollout schedule beyond Meta’s X statement. Separately, GitHub issue discussions around a Claude Code User-Agent privacy concern and a Hyperstition Hyperkernel open-source claim remain outside today’s confirmed stack until primary status and figures are clearer. Rumors of further GPT-6.2 Astra or Sol drops stay unofficial.