The last few weeks have been full of developments on the AI front. On the one hand, the establishment of the European Pact on AI has brought the AI Act and all the steps involved in putting it into action back into the spotlight. On the other hand, the news of the two behemoths OpenAI and TikTok is causing a stir, with a mixture of wonder, fear and bad buzz.
The European pact on AI, a first step towards the implementation of the AI Act?
An AI pact that commits to nothing
As we await the actual implementation of the AI Act in the summer of 2026, the EU is trying to unite companies around a European AI Pact. Voluntary companies are committing themselves around three axes:
- Implementing a corporate governance strategy,
- Identifying high-risk AIs,
- Promoting AI within the company.
Any organization can be a signatory to the Pact, following the example of Google, Microsoft and Cisco. While the Pact is in no way coercive, signatory companies are nevertheless publicly displaying a proactive approach to the framing of AI in their organization, and can take a small step ahead of the application of the AI Act.
These giants who have opted out of the Pact: Meta, Apple, TikTok...
More than 120 companies have signed up to the European AI Pact, but a number of big names are missing, including Meta, Apple, Mistral, Hugging Face and Tik Tok. What lessons can we learn from this?
There are two possible scenarios. The first focuses on the very nature of the AI Pact, which is rather superficial and uninviting. Companies such as Meta, Apple and Tik Tok are probably sufficiently advanced in the integration and supervision of AI.
The other possible analysis is that the "non-signature" of this pact is a message of protest from these giants, marking a lack of consensus on the content of the AI Act. This is because, unlike the AI Pact, the AI Act will have a real impact on these digital giants.
This could be seen as a way of opening up discussions with legislators, an idea reinforced by the departure of Thierry BRETON, the initiator of the AI Pact, which could be interpreted as a sign of a possible relaxation of European regulation.
France gives AI its place
Since September, France has had a Secretariat of State which, in addition to its traditional responsibility for digital, is now also dedicated to Artificial Intelligence. This is a national first, and augurs well for a stronger promotion of AI and digital technology in France. What's more, this Secretariat of State dedicated to Artificial Intelligence and Digital is attached to the Ministry of Higher Education and Research, and no longer to the Ministry of the Economy as was the former Secretary of State for Digital. This decision sends a positive message, symbolizing the French government's investment in innovation and R&D.
OpenAI still top of the bill
OpenAI wipes out its deficit with a fund-raising drive
Today, as the 3rd largest unlisted company in the world (behind SpaceX and ByteDance, parent company of Tik Tok), OpenAI has raised a further $6.6 billion for a total valuation of $157 billion. While the figures may seem staggering, this fund-raising is unlikely to be the company's last.
Unlike Google or Microsoft, who generate profits through software licenses or advertising, for OpenAI, AI is its sole source of revenue. So it's hard to train and invest in R&D without third-party services to help finance it.
With several million active users per week, OpenAI is today's most widely used solution. These figures reflect not only its success, but also the immense machine behind it, and the colossal costs required to ensure computing power. And that's without taking into account the R&D required to remain competitive on the market and to counter China, which is investing heavily in LLMs. Even if rumors of an increase in ChatGPT subscription fees prove true, it's a safe bet that the extra revenue alone won't be enough to cover the costs.
Missions and resignations at OpenAI
While the many changes within the OpenAI teams, such as the departure of Mira MURATI (Technical Director) and Ilya SUTSKEVER (co-founder), have raised fears about the company's future, this is without taking into account other announcements that point to a brighter future. The recruitment, for example, of Sebastian Bubeck (ex-VP AI at Microsoft, in charge of making LLM computation lighter) demonstrates OpenAI's determination to pursue its investments, attract talent and perhaps work in parallel on the compression and quantification of its models.
In addition, and in response to the criticism OpenAI is facing, Sam Altman has created a dedicated Security Committee from which he has voluntarily excluded himself. A signal sent out by OpenAI's management to reassure us of the safety of the models deployed by the firm.
OpenAI launches 2 new voice- and search-based services
Finally, OpenAI has struck again in recent weeks with the release of 2 new services:
- ChatGPT's "Voice to Voice": Since October 22 in France, ChatGPT's mobile application with its "Voice Advanced" mode has enabled users to start a real-time, latency-free voice conversation with OpenAI's assistant. It features a human-like voice, capable of understanding emotions, being interrupted mid-sentence, and even playing roles or accents.
- GPT Search: Last week OpenAI unveiled "GPT Search", turning its LLM into a real search engine. During my tests, where Gemini (Google's generative AI) failed to inform me, OpenAI synthesized information with remarkable precision, citing all its sources. The nth disruptive innovation in the field of generative AI from OpenAI.
Tik Tok shocks: 700 redundancies in Malaysia
Moderation thanks to Artificial Intelligence
From another giant, the announcement of the layoff of 700 employees in Malaysia is causing quite a stir. Tik Tok has decided to use AI to manage the moderation of its content. The dilemma of replacing man by machine is not new, and strikes once again. While the social and economic issues surrounding the loss of 700 jobs are obviously not to be underestimated, the impossibility of human moderation on the world's 5th largest social network, with 1.5 billion active users and no fewer than 34 million videos posted every day, is quite conceivable.
Entrusting content moderation to AI seems a logical enough solution, and one already applied by others such as Meta and Microsoft, in 2022 and 2023. Only AI can cope with the sheer scale of the task, enabling a large proportion of content to be moderated quickly and efficiently, leaving the more complex moderation cases to humans. While AI has the advantage of indefatigable efficiency, it is incapable of rapidly understanding the evolution of languages, irony, cultural references, trends and fashions. Similarly, in the event of a dispute over content moderation, it's a safe bet that AI will always be less relevant than humans. Rather, the challenge will be to find the balance between a hybrid model of AI moderation and human arbitration.
These dismissals probably also echo the regulatory pressure put on GAFAM by the Malaysian government to implement stricter frameworks in terms of cyber offenses.
These three news items are probably the most talked-about in recent weeks. Despite the fears that some of the announcements may have aroused, it's important to keep an eye on the impact that the various decisions will have, whether it's the organizational changes at OpenAI or the integration of AI for moderation at Tik Tok. In the time remaining before the AI Act becomes a reality, we'll be able to observe the repercussions, if any, of the Pact.
Jérôme Malzac
Innovation Director at Micropole
