Good afternoon John Shelley Bloomberg News.

I have a question regarding cybersecurity and AI, specifically ENISA's recently unveiled access to Anthropics Mythos model,

and I suppose the question is for Thomas.

I was essentially my first question is, First of all, why did it take so long?

Like 3 months is not necessarily a great achievement to Get access after 3 months, almost 3 months after the UK's AI Safety Institute did,

for instance, and also, Have we requested access to the actual latest Mythos,

which is Mythos 5.1? Thanks.

On the latter part of the question, nothing to announce on that front.

On the first part of the question, I would challenge you and rebut.

I would say that this is a great achievement for the European Union. It is a great achievement.

Our cybersecurity agency indeed now has been granted access to ENISA. Now why 3 months?

Precisely because we wanted to make sure the invitation came into the Commission quite a while ago.

Now what was the process in between,

we had constructive engagement with a company like Anthropic and we needed of course to put in place appropriate safeguards and terms and conditions for a safe access that has now been achieved and this materialized in us having now official access.

The good news should not be underestimated here, indeed, the fact that we have now official access. Rejoice. The joke was missed. We all rejoice.

Now, the first question is, OK, well, can you give us a vague idea what kind of,

Better terms, better protections, better safeguards you have managed to negotiate throughout these three months.

And the second is, I mean, you didn't comment, but I just wonder why you didn't,

you are not asking for access to what is, as of today, the most powerful and therefore most dangerous model.

I'm not saying that we didn't request access. I'm not confirming that we requested access.

I just said we now have access to the previous model, which is a very advanced and capable model, so we're focusing on 5 and that one now which we're currently testing,

I mean, our cybersecurity Agency is currently testing it, and Jan, please forgive me, but I forgot your first question.

Yes, what have we extracted, of course confidential discussions with Anthropic.

It's very difficult for me from this podium here to go into the details.

Just a little bit of gist that I can give you is that of course when you have access to such a model,

it is important for us not to expose potential vulnerabilities on our side.

So this is where we had to put in place a number of safeguards.

I will not go into the details of these,

but these were essentially the discussions we're having with anthropic and again good discussions that resulted in the positive outcome that we're having today. Good, thank you. Other questions to, Thomas today? Yes.

Hello, thank you, Maximilian Henning, for Euractiv on AI safety.

We already spoke on Monday about the news from the colleagues from Reuters about OpenAI agents getting involved with the European website.

There's now been further, revelations that there are several more, many more websites potentially affected by that. I want to ask for the latest. What does the Commission know there? What has OpenAI told you? What do you have? Thank you.

A lot of questions going into confidential discussions, but I can tell you, Maximilian, is that we're, of course, aware of the latest reporting which we have seen.

We're aware of the latest incidents and we're in close contact with OpenAI.

Now let me make one point extremely clear.

Is that it is high time for these providers to get their house in order and to make sure that they get their advanced model under control. The AI Act is fully enforced.

It's not just a set of rules on paper anymore. The Commission has enforcement powers.

We have already started to use them with the RFIs that were sent to several companies to make sure that they comply with their obligations.

And as a reminder, what can we do beyond these requests for information if it gets bad or worse,

we can definitely evaluate models.

We can require risk mitigation measures and in extreme cases and where necessary,

we can also restrict, withdraw, or even recall AI models. Thank you. Yes, please.

Apple confirmed on Wednesday that its new flagship assistant Siri AI won't launch,

won't be available when it launches in October in the EU on iPhones.

It will be available on computers, I think.

Can you give us an update on the discussions between the Commission and the company on the compatibility. AI with the DMA.

Let me start by saying that we are having constructive discussions currently taking place with Apple on the potential future offer of Siri in the EU.

Now let me make one point very clear.

The decision for now not to roll out Siri in the EU is Apple's decision because nothing in the DMA prevents any company from offering innovative products in the European Union.

Nothing in the DMA requires any company to lower its safety and security standards to the opposite.

The only thing that we're requesting is interoperability for our citizens who want to use Siri AI and the Apple product as it currently stands. They are free to do it.

For those who would like to use alternatives, that's where the interoperability criteria kicks in.

Apple needs to slightly open the market to make sure that competitors can compete on an equal foot.

And Razyie.