> This will result in Mozilla and Apple having to licence Google's model, or ship a model that's quirks-compatible with the Google model in order to be interoperable. It may also become difficult for Chrome to update its own model for the same reasons.
Google is again doing Evil.
I am very annoyed that Google kind of de-facto controls the www (through chrome, let's be honest here).
We really need to change this. I don't have a good solution here, but it can not continue that way.
I wonder if it makes sense for browser vendors to agree upon and ship various ‘standard models’ that are released into the public domain or something, and the API lets you pick between them.
The models themselves would be standardized and the weights and everything should be identical between browsers. They’d be standard and ‘web-safe’ like CSS colors or fonts. Probably would help to give them really boring/unbranded names too. These would work identically across browsers and web developers can rely on them existing on modern setups.
If you want more models, you could install them as a user or your browser could ship them or the web developers could bundle them through a CDN (and another standard for shared big files across domains would probably be needed)
Wow. I had no idea that people would misinterpret what I was saying in this way. I was not meaning to imply it was an expectation of users or developers. I was meaning it as a statement of what was currently a growing industry trend by OS and browser vendors, of shipping or preparing to ship LMs.
By now the statement could probably be amended from "expected to gain access to" to "shipping with".
I hope the team maintaining the project now makes such an update, since apparently it's confusing so many people!
AI massively empowers people who are incapable of anything except bikeshedding. It itself is very likely to be a bikeshed (but there are legitimate uses), and it also gives them to power to drone on until they overpower any opposition to their useless ideas.
Everything is increasingly expected to gain bikesheds.
If every browser vendor already has their experimental APIs that can work with different models, it might be a good idea to standardize this in WhatWG living standards (which would still be bad user experience on today's consumer hardware)
But if no browser other than Chrome supports this, and only Google's (proprietary) model (edit: plus Microsoft's Phi-4 mini in Edge), it should be clear it's Google abusing its position. There is nothing worth standardizing.
And we have seen that too many times -- FLoC/Privacy Sandbox/Topics API, Web Environment Integrity just to name a few. Google has been relentless in using its dominant position to push terrible ideas that harm both users and other browser vendors but help only Google's business.
Archibald is anti-AI. 70+% of his public statements have demonstrated that.
He is more or less aligned with the current most common sentiment in the west which is largely publicly against AI.
But realistically it's just slow adaptation, network effects, etc.
To give an example, before the MLB rolled out the Automated Ball Strike system this year, last year maybe 65+% of the sentiment in discussions about it was negative or in some cases just neutral.
Now that it has rolled out, 95% of the sentiment online about ABS is positive. The main comment by far is, why didn't they do this before, and why don't they do it automatically on all pitches now.
There are certain cognitive and informational flow limitations in society that will cause this to be delayed, just like all major technological advancements.
But once it rolls out, the perspective you hear online will be about digital sovereignty/personal data autonomy, now we aren't required to send our data to an external provider for AI, why wasn't this available before. People will probably assume it was blocked because it reduced a major source of data for advertising or something.
And overall AI and robotics in the future will be seen as the greatest enabling factor for increased equality in society.
It's really just this underlying dislike of and disrespect for technology that much of the western public has. Which may turn out to be one of the reasons that we lose our de facto leadership position in the world.
When I posted this, I linked to the latest statement https://github.com/mozilla/standards-positions/issues/1213#i..., which is the content relevant to the title (the details of our opposition to the API). Unfortunately someone removed the link to the specific post.
^ didnt realize who posted the opposition - this is Jake Archibald, a longtime googler on the Chrome team, now joining Mozilla and posting opposition to the Chrome API. no wonder the criticism is so well argued. most be a relief to not have to toe the party line on this one.
He's very familiar with standards-positions repos no doubt, and it reads like a typical defense against Google rolling something out without getting input. They don't suggest changes, they just try to throw out the whole thing. I think they might be hoping that if it ges thrown out, a collaborative effort will be formed to write if from scratch rather than start from the perspective of the Google Chrome team. I haven't seen it happen that way much, though, so I think it would be better if they just suggest specific changes to it rather than rejecting it.
I wonder if this is a generational thing of fresh young people that already cannot live without LLMs versus crusty old people that don’t want to require a super computer just to run a web browser that violates all their privacy.
To me this sounds like the point where people start looking at and developing alternatives to the browser/web.
A little off-topic, I honestly don't think it's as much as the browser interface that needs to be reworked as it is the idea of operating systems in general.
I don't know what the right answer is, but having used Niri/Wayland vs. GNOME vs. Windows vs. Mac... I will never go back to a non-tiling desktop and a none-kb driven workflow for desktop window management.
> According to Chrome's documentation, to use the prompt API you must 'acknowledge' Google's Generative AI Prohibited Uses Policy. Elements of this policy go beyond law. For example:
>> Do not engage … generating or distributing content that facilitates … Sexually explicit content
Do not engage in misinformation, misrepresentation, or misleading activities. This includes … Facilitating misleading claims related to governmental or democratic processes
> This seems like a bad direction for an API on the web platform, and sets a worrying precedent for more APIs that have UA-specific rules around usage.
I will say this more strongly—I think it is completely insane, and a violation of free expression principles, for a browser API to have content restrictions.
The nice thing about open protocols is that we don't have to endorse or use one implementation over another, yet, somehow, the browser monopoly continues to be a standing dilemma.
There are nice projects, like ungoogled chromium, tor, and many more, but I find the biggest issue is that there isn't a voice out there for the average person and a project that connects with the masses.
I think another issue is that a lot of the uninformed users have a strong apathy for the causes and ways the message is delivered, they rather engage and connect with things that are "fun" and want less friction rather than freedom and control.
How do we solve this? How do we make the browser ours, by the people, and for the people?
"A personal example: I created a system prompt for creating announcements for a home automation system. The Gemini model I was using initially responded in a very US-American way, which didn't fit the British voice of my speaker. I told the model, via the system prompt, that the output was being spoken in a British voice, but the result was a bad US-American impersonation of British ("a'waight guv'nor apples and pears" etc etc), so I had to iterate further to 'tone it down' and speak actual British.
In this process, the system prompt becomes tailored to the model. Other models will have different quirks. Things added to the system prompt for one model may be an overcorrection for another."
This API is a neutral party and can be iterated on. These objections are particular objections, to an implementation detail. One that can be swapped out, iterated on, improved on, changed, tailored to the user, by the user working with their user-agent and the LLM choice. One whose failure seems in the realm of personal preference rather than fundamental or harmful or damaging. So what if the English isn't exactly to your liking for the current model? To rest the argument against on such petty matters makes my head reel.
Iterating & improving on the base model & some system prompts allowed in this spec. Rather than using the API to try to cajole this very particular behavior that Jake seems to want everywhere, the user agent could let him set system prompts or pick a more suitable UK trained model, if that's really so important to them. The user agent is the proper channel for the agency Jake is seeking here. Theres nothing preventing the user & their user agent from negotiating what model they use. I don't think we all should be held hostage to naysaying by people who decide that the ability to have "the model needs to talk like a pirate, but this model didn't do that well" decision making. That's circumstantial nonsense, blocking the user agent from being able to work with the user to extend user agency, over such a narrow concern, that must be free to iterate anyhow!!
This API is the best basis we have to allow this negotiation to happen, out of band, outside the scope of the web API offered here, by agents. It's not up to the page to define this in the first place. It's the browser, the user agent that (as always has been the case on the web) builds user agency at its offered level of customizability & complexity. Maybe not every browser offers a "speak like a pirate". That lack is not a ding on the web prompt API!
The objections as stated have no resolution. This is a forever block for all time that is aggreived because not every model is going to behave exactly predictably perfectly. And there's no possible way out of this condundrum. The grievances of this submission are that sites will try to work around this, but the greivance here is built around the assumption that all agency has to lie with the site, that it's the site's obligation to fix US vs British English, that it's the site's need to tailor the agent.
That's not feasible not possible not sensible ever. The user agent is the mediating agent between the site and the user and the agent. That is going to be a complex evolving and dynamic relationship. The "failure" Jake cites here of the site to fully sculpt the experience is unreasonable, is an anti goal. It's up to the user and the browser to shape the agent for them, not each site.
I find these objections to be deeply deeply misguided. But worse, I find them to insist in perfection. There is no direction offered, no improvement suggested. The site can't make agents perfect therefore no one gets agents. That's all this says. It's fucking bullshit and fuck this a lot.
(I love Jake and they have done so much good so so so many times, but this is an impossible situation they are creating while leaving zero space for possibility for maybe and zero leadership for how else we can do what obviously must be done. Alas I think Mozilla at large had become the anti-possible company of web standards, which is a detestable position, one I had hoped might improve, eventually.)
Web API features should be things that are necessary to enable features in Web applications. We don't need the browser to have a Prompt API to enable web applications to have goofy chatbots lurking in the corner. WebDevs are perfectly capable of ruining their websites on their own.
I find this a weird discussion at the current point.
Shouldn't be there a basic process for allowing such an API as a alpha people can play around with and then there will be adjustments?
No one will start using this in production if they don't have a very good and specific use case. I mean you don't just run 2gb ML models in your browser today on a massive scale.
That discussion has a quote about querying the LLM for version information. If the models hallucinate/make up court citations, work and facts, what makes them believe that the model provided a genuine version number as opposed to an generatively constructed string?
So we can’t have XSLT fast and efficient templating syntax but Prompt APIs with potential attack injection vectors are cool as long as they’re generic enough for all megacorps to drop in? No security risks here huh? Not trying to increase the attack surface huh?
The Prompt API has some advantages like being a little simpler for some things and some potential to standardize a little bit more in some way, but it looks like from this that it will be delayed unfortunately.
However, WebLLM (a library, not actual Web API) https://github.com/mlc-ai/web-llm is more capable and will already work using WebGPU.
I was formerly the design lead / spec editor for this API while I worked at Google. I retired in 2025-09, before it got shipped. The following contains no inside knowledge.
I am sympathetic to all of Mozilla's concerns here, even though on balance I believe Chromium's decision to ship was the right one.
---
On interoperability, I agree that this is a tough case. But I am more optimistic than Mozilla that developers will use this API in a way that can work across different models.
First, they will be somewhat forced to, because Chrome will change the model over time. (It already changed from Gemini Nano 2 to 3, and I suspect it'll change to 4 soon if it hasn't already.) Edge is already shipping a Phi-based version. A small number of users are using other models via extensions like https://aibrow.ai/. And it's very possible Safari might join the party, exposing the Apple Foundation Models that ship with iOS via this API. (When the Foundation Models API came out, we were struck by how similar it was to the prompt API designs that preceded it, and were hopeful that Apple was going to do a surprise announcement of shipping the prompt API. It hasn't happened yet, but I still think it might soon.)
Second, we designed the API to steer developers in that direction as much as possible, e.g. encouraging the use of structured output constraints. There are also lots of clear error paths, that almost force developers to use this as a progressive enhancement. (E.g., the existence of low-memory/disk space devices.) So it's very unlikely we'll see developers build sites that are gated on this API existing. It'll mostly be used to sprinkle some AI magic, or let users do cool things without entering some cloud API keys.
I made similar arguments for the writing assistance APIs at [1]. As I said there, the prompt AI is trickier than the writing assistance APIs. But I believe it's a difference of degree, not kind. The web has many nondeterministic APIs that access some underlying part of the system, from geolocation to speech recognition/synthesis, all the way up to these AI-based ones. The question is where you draw the line. Mozilla seems to be giving some signals (not yet definite) that translation is on the OK side of the line, but summarization/writing/rewriting/prompting is not. That's a very reasonable position for them to take on behalf of their users. I imagine the Chromium project is hoping that over time, in-the-wild experience with these APIs shows that the benefits outweigh the risks and costs, and so Mozilla (and Apple) follow in shipping them as well. That's definitely happened in other cases, e.g., Mozilla recently indicating interest [2] in implementing WebBluetooth, WebHID, WebNFC, WebSerial, and WebUSB after years of taking a wait-and-see attitude.
You can learn more about my general thoughts on this question of shipping APIs first, and how the Chromium project takes on first-mover risks, at [3], which I wrote during my time on the Chrome team.
---
On the prohibited use policy, I agree that this is just absurd on Chrome's part. This is not how web APIs should work. It smacks of lawyers trying to throw something out there to cover themselves, or of corporate policy being set at the top level for "all AI uses" and then applied even for web APIs where that makes less sense.
The only saving grace is that I suspect it won't actually trigger. Because, as Mozilla points out, it's quite impractical to enforce. But it's still wrong.
I hope Chrome changes this, although I'm not holding my breath.
I did find it interesting that Gemma seems to have a similar terms of use [4]. (Open-weights, not open-source!) As do the Apple Foundation Models in iOS [5]. So unfortunately if the Chrome team were to push for a no-TOS API, they might be forging new ground, which is always difficult in a large company.
I’m kinda terrified by the security implications of the Prompt API.
This is a way for web services to make your computer complete large amounts of compute at their behest. Tokens have value. There will be incentive for bad actors to use your local LLM for their own purposes, much like hostile crypto mining payloads.
This is an obvious target for prompt injection attacks and other malicious remote code execution. In many ways, model prompts ARE programs. The browser / local device would need to provide an LLM with the same sandbox guarantees as the rest of the browser. Can they be trusted to do that? Does anyone understand this well enough to do that with confidence?
I’m a big fan of local models, but I would be very cautious about letting random websites call the model I’m hosting on my local machine with open source software.
1) This will be a new source of fingerprinting information and this is difficult to fake to fool fingerprinting scripts, so it can be abused for "device verification". There should be no ability to "verify" a browser, and anyone should be able to emulate any browser. This is the most important point, I thought Google people are smart enough to see it.
2) LLMs use lot of memory and CPU time, for many users they would slow down their system significantly, and given current RAM prices, upgrades are very expensive. If the website relies on local model, it would work slow on cheap devices.
3) The API seems to be tailored for specific LLM like OpenAI.
4) This can be used to push competitors who do not have an AI model from the browser market - the sites would break because they will be made with expectation of having Google Gemini model and would not work with other models. For example, the sites would break in national browsers not having an AI model. There should be no "first-class" and "second-class" browsers.
The explainer claims that this would allow the user to process the data locally without sending it anywhere. But why does Google Gemini local model have "Prohobited Use Policy" then? Why should they bother about prompts and responses they never learn about?
While offline LLM access seems like a good idea, the website could use WebGPU for this without building LLM into the browser (or they could improve WebGPU for better handling ML models). Or everyone should use the same, open source, LLM.
49 comments
[ 4.3 ms ] story [ 67.5 ms ] threadGoogle is again doing Evil.
I am very annoyed that Google kind of de-facto controls the www (through chrome, let's be honest here).
We really need to change this. I don't have a good solution here, but it can not continue that way.
The models themselves would be standardized and the weights and everything should be identical between browsers. They’d be standard and ‘web-safe’ like CSS colors or fonts. Probably would help to give them really boring/unbranded names too. These would work identically across browsers and web developers can rely on them existing on modern setups.
If you want more models, you could install them as a user or your browser could ship them or the web developers could bundle them through a CDN (and another standard for shared big files across domains would probably be needed)
Are they?
[0] https://github.com/webmachinelearning/prompt-api/blob/main/R...
Wow. I had no idea that people would misinterpret what I was saying in this way. I was not meaning to imply it was an expectation of users or developers. I was meaning it as a statement of what was currently a growing industry trend by OS and browser vendors, of shipping or preparing to ship LMs.
By now the statement could probably be amended from "expected to gain access to" to "shipping with".
I hope the team maintaining the project now makes such an update, since apparently it's confusing so many people!
Everything is increasingly expected to gain bikesheds.
Can't wait for the CVEs.
But if no browser other than Chrome supports this, and only Google's (proprietary) model (edit: plus Microsoft's Phi-4 mini in Edge), it should be clear it's Google abusing its position. There is nothing worth standardizing.
And we have seen that too many times -- FLoC/Privacy Sandbox/Topics API, Web Environment Integrity just to name a few. Google has been relentless in using its dominant position to push terrible ideas that harm both users and other browser vendors but help only Google's business.
Surprised this did not really come up in previous discussion in https://news.ycombinator.com/item?id=47917026
PS: looks like Google's fanboys have arrived. Someone better finds good counterarguments, especially technical ones, instead of just downvoting.
He is more or less aligned with the current most common sentiment in the west which is largely publicly against AI.
But realistically it's just slow adaptation, network effects, etc.
To give an example, before the MLB rolled out the Automated Ball Strike system this year, last year maybe 65+% of the sentiment in discussions about it was negative or in some cases just neutral.
Now that it has rolled out, 95% of the sentiment online about ABS is positive. The main comment by far is, why didn't they do this before, and why don't they do it automatically on all pitches now.
There are certain cognitive and informational flow limitations in society that will cause this to be delayed, just like all major technological advancements.
But once it rolls out, the perspective you hear online will be about digital sovereignty/personal data autonomy, now we aren't required to send our data to an external provider for AI, why wasn't this available before. People will probably assume it was blocked because it reduced a major source of data for advertising or something.
And overall AI and robotics in the future will be seen as the greatest enabling factor for increased equality in society.
It's really just this underlying dislike of and disrespect for technology that much of the western public has. Which may turn out to be one of the reasons that we lose our de facto leadership position in the world.
To me this sounds like the point where people start looking at and developing alternatives to the browser/web.
It’s them articulating clear and logical reasons why the proposed API, in its current state, is bad for web interoperability.
That shipped sailed in 2008.
I don't know what the right answer is, but having used Niri/Wayland vs. GNOME vs. Windows vs. Mac... I will never go back to a non-tiling desktop and a none-kb driven workflow for desktop window management.
>> Do not engage … generating or distributing content that facilitates … Sexually explicit content Do not engage in misinformation, misrepresentation, or misleading activities. This includes … Facilitating misleading claims related to governmental or democratic processes
> This seems like a bad direction for an API on the web platform, and sets a worrying precedent for more APIs that have UA-specific rules around usage.
I will say this more strongly—I think it is completely insane, and a violation of free expression principles, for a browser API to have content restrictions.
There are nice projects, like ungoogled chromium, tor, and many more, but I find the biggest issue is that there isn't a voice out there for the average person and a project that connects with the masses.
I think another issue is that a lot of the uninformed users have a strong apathy for the causes and ways the message is delivered, they rather engage and connect with things that are "fun" and want less friction rather than freedom and control.
How do we solve this? How do we make the browser ours, by the people, and for the people?
Sorry, I'm just sad whenever I think of this.
[0]. https://developer.mozilla.org/en-US/docs/Web/HTML/Reference/...
From https://github.com/mozilla/standards-positions/issues/1213 :
"A personal example: I created a system prompt for creating announcements for a home automation system. The Gemini model I was using initially responded in a very US-American way, which didn't fit the British voice of my speaker. I told the model, via the system prompt, that the output was being spoken in a British voice, but the result was a bad US-American impersonation of British ("a'waight guv'nor apples and pears" etc etc), so I had to iterate further to 'tone it down' and speak actual British.
In this process, the system prompt becomes tailored to the model. Other models will have different quirks. Things added to the system prompt for one model may be an overcorrection for another."
This API is a neutral party and can be iterated on. These objections are particular objections, to an implementation detail. One that can be swapped out, iterated on, improved on, changed, tailored to the user, by the user working with their user-agent and the LLM choice. One whose failure seems in the realm of personal preference rather than fundamental or harmful or damaging. So what if the English isn't exactly to your liking for the current model? To rest the argument against on such petty matters makes my head reel.
Iterating & improving on the base model & some system prompts allowed in this spec. Rather than using the API to try to cajole this very particular behavior that Jake seems to want everywhere, the user agent could let him set system prompts or pick a more suitable UK trained model, if that's really so important to them. The user agent is the proper channel for the agency Jake is seeking here. Theres nothing preventing the user & their user agent from negotiating what model they use. I don't think we all should be held hostage to naysaying by people who decide that the ability to have "the model needs to talk like a pirate, but this model didn't do that well" decision making. That's circumstantial nonsense, blocking the user agent from being able to work with the user to extend user agency, over such a narrow concern, that must be free to iterate anyhow!!
This API is the best basis we have to allow this negotiation to happen, out of band, outside the scope of the web API offered here, by agents. It's not up to the page to define this in the first place. It's the browser, the user agent that (as always has been the case on the web) builds user agency at its offered level of customizability & complexity. Maybe not every browser offers a "speak like a pirate". That lack is not a ding on the web prompt API!
The objections as stated have no resolution. This is a forever block for all time that is aggreived because not every model is going to behave exactly predictably perfectly. And there's no possible way out of this condundrum. The grievances of this submission are that sites will try to work around this, but the greivance here is built around the assumption that all agency has to lie with the site, that it's the site's obligation to fix US vs British English, that it's the site's need to tailor the agent.
That's not feasible not possible not sensible ever. The user agent is the mediating agent between the site and the user and the agent. That is going to be a complex evolving and dynamic relationship. The "failure" Jake cites here of the site to fully sculpt the experience is unreasonable, is an anti goal. It's up to the user and the browser to shape the agent for them, not each site.
I find these objections to be deeply deeply misguided. But worse, I find them to insist in perfection. There is no direction offered, no improvement suggested. The site can't make agents perfect therefore no one gets agents. That's all this says. It's fucking bullshit and fuck this a lot.
(I love Jake and they have done so much good so so so many times, but this is an impossible situation they are creating while leaving zero space for possibility for maybe and zero leadership for how else we can do what obviously must be done. Alas I think Mozilla at large had become the anti-possible company of web standards, which is a detestable position, one I had hoped might improve, eventually.)
But most importantly this would enable us to finally write JavaScript like this:
const a = prompt("how much is 31c in Fahrenheit")
The future looks bright!
Shouldn't be there a basic process for allowing such an API as a alpha people can play around with and then there will be adjustments?
No one will start using this in production if they don't have a very good and specific use case. I mean you don't just run 2gb ML models in your browser today on a massive scale.
However, WebLLM (a library, not actual Web API) https://github.com/mlc-ai/web-llm is more capable and will already work using WebGPU.
I am sympathetic to all of Mozilla's concerns here, even though on balance I believe Chromium's decision to ship was the right one.
---
On interoperability, I agree that this is a tough case. But I am more optimistic than Mozilla that developers will use this API in a way that can work across different models.
First, they will be somewhat forced to, because Chrome will change the model over time. (It already changed from Gemini Nano 2 to 3, and I suspect it'll change to 4 soon if it hasn't already.) Edge is already shipping a Phi-based version. A small number of users are using other models via extensions like https://aibrow.ai/. And it's very possible Safari might join the party, exposing the Apple Foundation Models that ship with iOS via this API. (When the Foundation Models API came out, we were struck by how similar it was to the prompt API designs that preceded it, and were hopeful that Apple was going to do a surprise announcement of shipping the prompt API. It hasn't happened yet, but I still think it might soon.)
Second, we designed the API to steer developers in that direction as much as possible, e.g. encouraging the use of structured output constraints. There are also lots of clear error paths, that almost force developers to use this as a progressive enhancement. (E.g., the existence of low-memory/disk space devices.) So it's very unlikely we'll see developers build sites that are gated on this API existing. It'll mostly be used to sprinkle some AI magic, or let users do cool things without entering some cloud API keys.
I made similar arguments for the writing assistance APIs at [1]. As I said there, the prompt AI is trickier than the writing assistance APIs. But I believe it's a difference of degree, not kind. The web has many nondeterministic APIs that access some underlying part of the system, from geolocation to speech recognition/synthesis, all the way up to these AI-based ones. The question is where you draw the line. Mozilla seems to be giving some signals (not yet definite) that translation is on the OK side of the line, but summarization/writing/rewriting/prompting is not. That's a very reasonable position for them to take on behalf of their users. I imagine the Chromium project is hoping that over time, in-the-wild experience with these APIs shows that the benefits outweigh the risks and costs, and so Mozilla (and Apple) follow in shipping them as well. That's definitely happened in other cases, e.g., Mozilla recently indicating interest [2] in implementing WebBluetooth, WebHID, WebNFC, WebSerial, and WebUSB after years of taking a wait-and-see attitude.
You can learn more about my general thoughts on this question of shipping APIs first, and how the Chromium project takes on first-mover risks, at [3], which I wrote during my time on the Chrome team.
---
On the prohibited use policy, I agree that this is just absurd on Chrome's part. This is not how web APIs should work. It smacks of lawyers trying to throw something out there to cover themselves, or of corporate policy being set at the top level for "all AI uses" and then applied even for web APIs where that makes less sense.
The only saving grace is that I suspect it won't actually trigger. Because, as Mozilla points out, it's quite impractical to enforce. But it's still wrong.
I hope Chrome changes this, although I'm not holding my breath.
I did find it interesting that Gemma seems to have a similar terms of use [4]. (Open-weights, not open-source!) As do the Apple Foundation Models in iOS [5]. So unfortunately if the Chrome team were to push for a no-TOS API, they might be forging new ground, which is always difficult in a large company.
---...
This is a way for web services to make your computer complete large amounts of compute at their behest. Tokens have value. There will be incentive for bad actors to use your local LLM for their own purposes, much like hostile crypto mining payloads.
This is an obvious target for prompt injection attacks and other malicious remote code execution. In many ways, model prompts ARE programs. The browser / local device would need to provide an LLM with the same sandbox guarantees as the rest of the browser. Can they be trusted to do that? Does anyone understand this well enough to do that with confidence?
I’m a big fan of local models, but I would be very cautious about letting random websites call the model I’m hosting on my local machine with open source software.
1) This will be a new source of fingerprinting information and this is difficult to fake to fool fingerprinting scripts, so it can be abused for "device verification". There should be no ability to "verify" a browser, and anyone should be able to emulate any browser. This is the most important point, I thought Google people are smart enough to see it.
2) LLMs use lot of memory and CPU time, for many users they would slow down their system significantly, and given current RAM prices, upgrades are very expensive. If the website relies on local model, it would work slow on cheap devices.
3) The API seems to be tailored for specific LLM like OpenAI.
4) This can be used to push competitors who do not have an AI model from the browser market - the sites would break because they will be made with expectation of having Google Gemini model and would not work with other models. For example, the sites would break in national browsers not having an AI model. There should be no "first-class" and "second-class" browsers.
The explainer claims that this would allow the user to process the data locally without sending it anywhere. But why does Google Gemini local model have "Prohobited Use Policy" then? Why should they bother about prompts and responses they never learn about?
While offline LLM access seems like a good idea, the website could use WebGPU for this without building LLM into the browser (or they could improve WebGPU for better handling ML models). Or everyone should use the same, open source, LLM.