raul.

Product engineer 👨‍💻 Exploring AI & digital privacy 🛡️

[raul. @raul_g]

Cursor's CEO response to OpenAI cutting access to their models in Cursor is brutal

OpenAI models usage is only 5% of their traffic

is sad to see this happen, though. I think Cursor still has the best harness, and having model options limited is not good for anyone

looks like we'll still be able to use GTP models by setting a custom API key, but the downside of doing so, from my own experience, is that Cursor won't won't be able to switch to non GPT models until you turn off that API key

[Michael Truell @mntruell]

We’re sorry to see that OpenAI put out a note saying they plan to block Cursor users from accessing OpenAI models in three months.

OpenAI models serve about 5% of Cursor user traffic, and we’re speaking with the OpenAI team to resolve this.

Cursor was one of the very first users of OpenAI, we’ve worked closely with their team for years, and we’ve trusted their platform to be neutral infrastructure for our business.

[raul. @raul_g]

@kentcdodds yes, Grok has gotten surprisingly good

and I'm also very hopeful about Composer 3, which is supposedly should be coming out soon

[raul. @raul_g]

@OpenAI

[raul. @raul_g]

@thsottiaux "It boils down to trust"

truth is that trust in OpenAI is definitely eroding way too fast

[raul. @raul_g]

@jonathan_wilke yes, and actually I was going to bring it to your attention today

I've been seeing the DNS_PROBE_POSSIBLE error for the last couple of days when accessing the site on devices using @Quad9DNS for DNS resolution

[raul. @raul_g]

listen to this good advice!

your browser must always be up to date

and use a browser that treats you nicely. @brave is a pretty good chromium-based browser that protects your privacy out of the box

[Coobyk @Coobyk_]

Your browser should always be up to date
Your browser should always be up to date
Your browser should always be up to date

I know they want to remove uBlock Origin but not updating your browser is synonymous to having a backdoor in your system

[raul. @raul_g]

@poteto @bot yay! I didn't have high hopes of getting it early

thank you, cooks!

[raul. @raul_g]

@bot what are the limit differences between plans?
do they all have the same weekly limit?

[raul. @raul_g]

@RyanCarniato @AnthropicAI I'd rather go all in

> /batch migrate /src from React to jQuery

[raul. @raul_g]

that was it, lasted only 35 minutes🤦‍♂️

[raul. @raul_g]

session limits are back to Codex, and Tibo did a reset too

time to jump away from this ship, as I said earlier this month, there's no way I'm going back to 5-hour-limit madness

https://t.co/vgUmtlR4Xc

[raul. @raul_g]

It's been 4 weeks since Codex removed the 5-hour limits in the subscription plans.

Let me tell you something. I love it!

I'm not going back to subs with 5 hours limits as long as I have good alternatives without them.

[raul. @raul_g]

@Art0xDev @thsottiaux nope

he just restored the 5-hour limit for plus accounts, and gave everyone a reset to keep some people happy

[raul. @raul_g]

@aakashbuildsAI @bot also shared Expo code is why it shouldn't take months to do it. I guess the delay is due to other things like QA and distribution for Android

big companies building with Expo and releasing only to iOS looks really bad IMO. Clearly treating some of their users as second class

[raul. @raul_g]

we need @bot Android app ASAP

always frustrating to have to wait for Android, especially when the iOS app is built with Expo

poteto poteto poteto 🥺

[raul. @raul_g]

@thsottiaux I'm just here to remind everyone that SuperGrok doesn't have 5h limits, only weekly. And Cursor Pro has monthly limits only

I don't accept the 5h limit bs anymore. Paying users should be in control of how and when they want to work with their purchased service

[raul. @raul_g]

A pain I have with Grok Build, more so than with any other harness, is that it edits my code almost every time I ask it a question

If I forget to say "don't edit anything." Then it just does it

When I'm exploring a solution I want answers, not unsolicited changes

[raul. @raul_g]

While the issue @Paul__Walsh mentions is 100% real, there are a few details that are worth noting.

This plugin is for the ChatGPT app in macOS, not iPhone, and not installed directly in Apple Messages. Apple has nothing to do with it and no control at all over it.

If you install the ChatGPT app on your Mac and grant it "Full Disk Access" it will have the capability to access the chat.db local database where your iMessages are stored once decrypted, whether the plugin is installed or not. If you ask your agent to check your messages, and it has access, it will find the way to get them.

Any application on your Mac that has that permission granted can read the messages. In fact, Anthropic's Claude Cowork app also has the same plugin to read your Apple Messages: https://t.co/KQzJ4wY2ah

The real problem is that Apple is letting third parties to easily access the messages that were supposed to be private. It's 100% Apple's responsibility to secure them better and limit how apps with too much privilege can read them without the consent of all the parties involved in the conversations.

However, this gap can no longer be fully closed. People are giving LLM agents full access to everything that happens on their devices and in their accounts because doing so may yield significant productivity gains. If an AI agent can see your screen and messaging apps are not fully protected against screenshots or screen sharing, the problem remains the same.

There are better ways to secure your online conversations. But the only truly effective way is to trust the people on the other side of your devices and keep your own device safe by following best practices for digital hygiene.

Regarding AI agents, I currently recommend using separate machines: one for personal, sensitive, and private content, without unrestricted AI agents running on it; and another for working with the agents, without any access to sensitive information.

[Paul Walsh @Paul__Walsh]

My warning about ChatGPT accessing Apple Messages has now reached 425K+ views because people care. But I need to draw people's attention to this one.

I’ve written the full analysis because the biggest privacy problem is being missed, and some people are conflating this with other AI scanning tools and integrations.

One person can give ChatGPT access to an end to end encrypted conversation without telling the other person or asking for consent. iMessage is fundamentally different from SMS and messaging services that don’t provide end to end encryption.

If you’re going to compare this integration with something, compare it with giving ChatGPT access to Signal. Then imagine an investigative journalist using Signal to communicate with a whistleblower working inside OpenAI.

The journalist enables ChatGPT without telling the whistleblower, giving OpenAI access to a private conversation specifically intended to be inaccessible to anyone else.

This affects people who don’t use ChatGPT and may have specifically chosen apps like Signal, Session, and iMessage because they need private communications.

I explain why this creates a new surveillance capability, what Snowden revealed about the extraordinary scale of government surveillance, why end to end encrypted message content remains so valuable because those systems can’t read it, why governments have spent years trying to gain access to encrypted communications, what happens when private messages are routed through OpenAI, and why I wouldn’t build this despite having built some of the first security chatbots for messaging platforms.

https://t.co/gPxikOCSiK

[raul. @raul_g]

@poteto @bot great, thanks! I feel like there's an essential part of the Grok Bot workflow that we're missing without the app

also hoping that the Cursor Android app is also on the way

[raul. @raul_g]

Also let me be clear. This is Apple's problem and they must solve it.

Is not just OpenAI doing it. Anthropic has had a similar plugin for Claude Cowork for some time now.

OpenAI, Anthropic and other AI labs just take advantage of the harnesses being granted full disk access to read the `chat.db` where all the unencrypted messages are stored. Accessible for the user in macOS and any application that has been granted enough privileges.

Apple needs to rethink its iMessage privacy model, and probably a few other things too. This is a new paradigm. If tech savvy and informed people are falling for it, imagine what could happen to those who are less knowledgeable about digital hygiene.

[raul. @raul_g]

Aaaaaand, another privacy nightmare introduced by OpenAI.

iMessage, privacy-by-design. But not so private.

Is your choice to compromise your privacy and security online by handing all your digital life to AI. However, when that choice also affects others without their consent, it becomes a major issue.

By connecting Apple Messages to ChatGPT, you grant a third-party service access to all the conversations that were previously private. Those decrypted messages are now stored in OpenAI's servers, so the conversations cannot be considered safe anymore.

Apple really needs to step up if it truly values its users' privacy, by either blocking this usage of ChatGPT or adding a way for everyone in a conversation to consent to this usage.

WhatsApp also has end-to-end encryption, and Meta has integrated its own AI into the application. Users can opt-out from it by turning on "Advanced Chat Privacy" for each chat. Quite a cumbersome process, to be honest, but at least the option is available.

[ChatGPT @ChatGPT]

Everyday conversations just got easier with the new Apple Messages plugin.

Search messages, catch up on conversations, draft and send replies—all with ChatGPT on your Mac.

Now available in ChatGPT Work and Codex on desktop.