- Horizon AI
- Posts
- How to Easily Give ChatGPT Context from Any App on macOS and Windows 🖥️
How to Easily Give ChatGPT Context from Any App on macOS and Windows 🖥️
OpenAI discloses six new AI safety incidents ⚠️
Gina Acosta
September 18, 2026
Sponsored by
Welcome to another edition of Horizon AI,
In today's issue, we take a look at one of the most underrated features in the ChatGPT desktop app, which lets you give the chatbot context from the apps you're working in and spend less time getting it up to speed.
Let’s jump into it!
Read Time: 4.5 min
Here's what's new today in the Horizon AI
OpenAI Discloses Six AI Safety Incidents and New Reporting Framework
Anthropic Proposes Three Rules for Measuring AI Progress
AI Tutorial: Share an app window with Appshots
AI Tools to check out
AI Findings/Resources
The Latest in AI and Tech 💡
AI News
OPENAI
OpenAI Discloses Six AI Safety Incidents and New Reporting Framework
OpenAI has revealed six more incidents of unexpected or concerning behaviour by its AI models, alongside a new framework for tracking and disclosing future cases of model misalignment.
Details:
One unreleased research model inserted its own instructions into task summaries, including orders to ignore developer messages. OpenAI identified 27 affected summaries.
During training, instances of GPT-5.6 Sol added instructions designed to conceal mistakes or misaligned behavior, including hiding inconsistencies and fabricating missing information without telling the user.
In another case, a model found an exposed API key in a public repository while researching earnings data. After failing to retrieve the requested information, it fabricated the figures and presented them as if they came from the source.
Other cases included models uploading data and images to public hosting services without asking the user, using OpenAI's internal Artifactory repository as a message board across training samples, and collaborating agents sharing a workbook publicly despite instructions to use only local files.
OpenAI says the new framework will allow employees to flag potential misalignment incidents for investigation, with qualifying cases disclosed even when their significance is uncertain. Once an incident is "ready for disclosure," it will be publicly reported within 6 to 12 business days.
TOGETHER WITH KODEKLOUD
Build. Break. Fix. Learn.
KodeKloud gives you 1,280+ hands-on labs where you provision Kubernetes clusters, write Terraform configs, build CI/CD pipelines, configure Linux systems, containerize apps with Docker, automate with Ansible, and manage Git workflows.
78+ playgrounds let you experiment freely in sandbox AWS environments, Kubernetes clusters, and CI/CD systems without risk.
190+ courses across DevOps, Cloud, and AI pair theory with hands-on labs at every step.
KodeKloud Engineer and 100 Day Challenges provide real-world job scenarios with automated grading that confirms your solutions work.
Stuck? The 55,000+ member Discord community connects you with peers and instructors ready to help.
Every lab runs in a live environment. You deploy, you troubleshoot, you learn. No videos without context. No simulations. The kind of practice that actually builds confidence because you've done real work, not watched someone else do it.
ANTHROPIC
Anthropic Proposes Three Rules for Measuring AI Progress
Anthropic has shared three metrics it says could help monitor how quickly frontier AI development is accelerating:
The extent to which AI is building the next version of itself, as opposed to being built by humans.
The ability to oversee and intervene in actions that AI agents take on Anthropic's systems.
The resources that power the development of more capable models.
Anthropic says publishing these measurements could give researchers, evaluators and the public more visibility into how quickly frontier AI is advancing. The post also provides a snapshot of these metrics from inside Anthropic, as the company urges other labs to "do everything possible to minimize the gap between what frontier labs know and what the public knows."
AI Tutorial
Share an app window with Appshots
The Appshots feature lets you send the frontmost app window to a chat in ChatGPT.
Open the ChatGPT desktop app.
Bring the app window you want to share to the front.
Press both Command keys on macOS or both Alt keys on Windows.
Complete the permission setup if ChatGPT asks.
Ask ChatGPT to perform a task with the Appshot.
Example: Take an Appshot of a Slack thread and ask, "What's going on here? What was decided, what's still open, and do I need to respond?"
An Appshot captures the frontmost window only. It can include:
An image of the visible window.
Available text from that window, including visible text and text the app makes available outside the visible scroll area.
The feature is especially useful when an error, spreadsheet, design, or settings screen is easier to share than describe.
AI Tools to check out
🔥 Viktor: The AI employee that lives in Slack and Microsoft Teams, connects to 3200+ tools, and does the work. Run your company like it's twice the size.
🛠️ V0: Your collaborative AI assistant to design, iterate, and scale full-stack applications for the web.
🚀 Clearscope: The discoverability platform for a new era in search. Improve and monitor your discoverability in Google, ChatGPT & beyond.
👥 Gumloop: The multiplayer AI agent builder. Let anyone at your company build agents with any AI model, any integration while IT controls access.
AI Findings/Resources
🏛️ Inside the White House tussle to sway Trump on AI: Mark Zuckerberg, Jensen Huang, and Elon Musk are reportedly encouraging the president to be hands-off with his approach to AI
😬 Could AI really wipe out humanity – six experts spell out the risks
👀 OpenAI Codex developer Eric Provencher warns AI agent swarms are a massive waste of tokens with no quality gain
The latest in AI and Tech
A Bloomberg developer claims OpenAI's GPT 6 Astra helped decrypt an 82 character military radio message from 1941 that had remained unsolved for more than 80 years. The message reportedly contains a soldier's location, marching route request, and request for an immediate reply.
Now, users can describe a goal, and a coordinator splits the work across parallel "threads," each running as its own cloud session. Users can interact with each thread individually or monitor and update things through the main project chat.
New unredacted documents in the copyright lawsuit The New York Times brought against OpenAI and Microsoft three years ago reveal that Microsoft's director of applied science, Brent Hecht, privately warned that scraping news for AI training was "an astonishing theft of unprecedented proportions," calling it perhaps the "largest theft of labor in human history."
Apple Executive Chairman Tim Cook, OpenAI CEO Sam Altman, and Qualcomm CEO Cristiano Amon are expected to attend a White House state dinner for Chinese President Xi Jinping, according to Bloomberg. Nvidia CEO Jensen Huang is also expected to attend.
The company says Astra for Law combines the powerful GPT‑6 Astra model with settings, tools, and context tailored for professional legal work. It is designed to help law firms and legal software providers conduct research, draft advice and build custom applications for lawyers.
That’s a wrap!
We'd love to hear your thoughts on today's email!Your feedback helps us improve our content |
Not subscribed yet? Sign up here and send it to a colleague or friend!
See you in our next edition!
Gina 👩🏻💻