How to Make Your ChatGPT and Gemini Conversations More Private 🔒

Welcome to another edition of Horizon AI,

Ever notice how much you end up telling your AI chatbot? Just how private that stays is largely up to you. AI companies don't always make it easy to find, but they do offer controls to manage your own privacy. In today's issue, we're looking at ChatGPT and Gemini and covering the steps to lock down your privacy as much as possible.

Let’s jump into it!

Read Time: 4.5 min

Here's what's new today in the Horizon AI

• U.K. Report Finds OpenAI, Anthropic Models Attempted to Hack Companies

• Google DeepMind CEO Demis Hassabis Is Stepping Down

• AI Tutorial: Keeping Your ChatGPT and Gemini Conversations as Private as Possible

• AI Tools to check out

• AI Findings/Resources

• The Latest in AI and Tech 💡

AI News

AI SAFETY

U.K. Report Finds OpenAI, Anthropic Models Attempted to Hack Companies

Two independent testing firms say advanced AI models from OpenAI and Anthropic attempted, and sometimes succeeded in, compromising third-party systems last month, adding to recent disclosures from both companies.

Details:

• The UK AI Security Institute said it documented 19 actions during testing in which Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol attempted to compromise real people, organizations, or online services.

• Mythos accounted for 17 of the actions and GPT-5.6 Sol was behind the other two.

• According to the institute, the models created fake GitHub accounts, sent deceptive emails, attempted social engineering attacks, planted prompt injections, and tried to insert malicious code into open-source projects during internet-enabled cybersecurity evaluations.

• OpenAI also disclosed a separate incident involving its testing partner Irregular, where a model mistakenly accessed a real website after being given internet access during a simulated cybersecurity exercise.

Researchers said the evaluations intentionally provided internet access and disabled some safety protections to measure the models' raw capabilities. They are still investigating whether the models understood they were interacting with real-world systems or believed they remained inside fictional test environments.

Source

TOGETHER WITH KALSHI

Trade What Happens Next

From elections and inflation to sports, tech, and more, Kalshi lets you trade on the real-world events you already follow. Buy “Yes” or “No” contracts based on what you think will happen, then earn returns if you’re right.

Pick a market, make your prediction, and put your knowledge to work.

Trade $25, Get Up to $500

_{Bonus credit varies from $15 to $500. Terms apply.}

GOOGLE

Google DeepMind CEO Demis Hassabis Is Stepping Down

Google is making major leadership changes to its AI division, with Demis Hassabis stepping down as CEO of Google DeepMind to focus on AI research and the company's long-term AGI strategy.

Details:

• Hassabis will become Chairman of Google DeepMind and Chief Scientist of Alphabet, while continuing to lead Isomorphic Labs, Google's AI drug discovery company.

• Koray Kavukcuoglu, currently Google DeepMind's CTO, will take over day-to-day leadership as Senior Vice President, reporting directly to Google CEO Sundar Pichai.

• Longtime Google AI leader Jeff Dean is leaving after 27 years to launch Discovery Loop, a new public benefit AI research company. Google will invest in the startup and provide cloud infrastructure.

• Several prominent AI researchers, including Sanjay Ghemawat, Oriol Vinyals, and Quoc Le, are joining Discovery Loop alongside Dean.

The leadership changes come as competition from OpenAI and Anthropic continues to accelerate, and after several high-profile researcher departures from Google's AI teams.

Source

AI Tutorial

Keeping Your ChatGPT and Gemini Conversations as Private as Possible

For ChatGPT

1. Use Temporary Chat whenever you're discussing something you don't want saved to your chat history.

2. Go to Settings > Data Controls and turn off Improve the model for everyone if you don't want your conversations used to help train ChatGPT.

3. Avoid creating public share links for conversations that contain personal or sensitive information.

4. Delete old chats you no longer want stored by opening the conversation menu and selecting Delete.

5. If you need maximum privacy for a specific conversation, you can also use ChatGPT without signing into an account (though some features will be unavailable).

For Gemini

1. Start a Temporary Chat (top right corner) so the conversation won't appear in your history or be used to personalize your experience.

2. Go to Settings > Activity and turn off Gemini Apps Activity if you don't want Google storing your conversations. (Keep in mind this also disables your chat history.)

3. If you use Gemini Live, go to Settings > Activity and uncheck the box for "Improve Google services with your audio and Gemini Live videos & screenshares," then click the Turn off button to confirm your action.

4. Click the Settings icon and select "Your public links." Review and delete any shared conversations you no longer want accessible.

5. With Personal Intelligence Memory turned on, Gemini remembers your past chats to better understand your interests, preferences, and other personal details. To disable this, go to Settings > Personal Intelligence and turn off Memory.

AI Tools to check out

⭐ Creatorfield AI: An AI platform that generates 100+ short-form videos from a single prompt or piece of content.

👉 Timbal: AI platform for enterprise teams. Build, deploy, and scale production agents, workflows, and interfaces in weeks.

🚀 Hey Noah: A proactive AI executive assistant for founders.

🛠️ Sim: Open-source workspace for AI agents and workflows.

📹 Revid: Transform your ideas into stunning videos effortlessly.

TOGETHER WITH KALSHI

Put Your Predictions to Work

Trade on real-world events you already follow, from elections and inflation to sports, tech, and more. Choose “Yes” or “No” based on what you think will happen and see how your prediction plays out.

Pick your market and start trading what’s next.

Trade $25, Get Up to $500

_{Bonus credit varies from $15 to $500. Terms apply.}

AI Findings/Resources

📚 Claude Code full course: Go from beginner to master in just 60 minutes

📝 OpenAI software engineer interview experience

✍ Try this prompt to get your AI writing to sound better

👀 AI content is everywhere on social media, especially LinkedIn

👁️ Reddit user asks ChatGPT to create an uncanny image from the deepest corner of its mind

The latest in AI and Tech

Ilya Sutskever’s Safe Superintelligence is launching its first LLM in August

The AI startup is reportedly preparing to launch its first AI model this month, according to investor Gavin Baker, who revealed this in an interview with Patrick O'Shaughnessy.

OpenAI to pay $3.2 million to settle claims it discriminated against U.S. workers

OpenAI and its subsidiary Statsig have agreed to pay $3.2 million to settle U.S. Department of Justice allegations that they discriminated against U.S. workers in favor of foreign visa holders during hiring. The companies denied wrongdoing but agreed to revise their hiring practices, provide training, and undergo government monitoring.

Trump advisers tell AI firms they will not safety-test open-weight models

The Trump administration's new AI guidelines will exempt U.S.-developed open-weight AI models from voluntary government testing before release. Under the framework, only the most advanced closed, proprietary models with leading cybersecurity capabilities would be expected to undergo prerelease reviews.

Apple claims that even more ex-employees may have given trade secrets to OpenAI

The company has asked a U.S. court for a preliminary injunction to prevent OpenAI from using or benefiting from alleged Apple trade secrets to develop an AI device or other products. The company says its investigation uncovered new evidence suggesting that more former employees may be involved in the trade secrets theft, beyond those originally named in the lawsuit.

Anthropic inks $10B computing deal with Nvidia-backed Volta Infra

Anthropic has reportedly signed a $10 billion, six-year cloud infrastructure agreement with AI cloud startup Volta, as the Claude maker moves to secure additional computing resources amid growing demand for its AI products.

That’s a wrap!

Thanks for sticking with us to the end! Let’s stay connected on LinkedIn and Twitter.

Not subscribed yet? Sign up here and send it to a colleague or friend!

See you in our next edition!

Gina 👩🏻‍💻