Claude Fable 5.2, Paris AI appeal, Instagram age checks

Four developments are worth watching on Tuesday (Nov. 10) for anyone in Brazil who uses or buys AI services. Anthropic released Claude Fable 5.2, which the company says has been available on the three largest cloud platforms since Monday (Nov. 9). Meta is rolling out AI age estimation for Instagram users worldwide. In Paris, governments and companies signed onto a declaration on the safety of the most advanced AI models, with no Latin American head of government among those attending. And Google unveiled new AI chip systems for cloud customers.

Anthropic: Fable 5.2 arrives on the major clouds

In a Nov. 9 announcement, Anthropic introduced Claude Fable 5.2 and Claude Mythos 5.2. Both are the same underlying model, Anthropic says. The difference lies in the safeguards: Fable carries stronger protections against misuse in cybersecurity and biology, which Anthropic says lets it offer the model to all users. Mythos stays limited to organizations that apply through verification programs in the life sciences and cybersecurity fields.

For those footing the bill, Anthropic says Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, which it describes as lower than the 5.1 models’ pricing. An optional fast mode speeds up generation by up to 2.5 times, at $16 and $80 per million tokens, respectively. According to the company, fast mode is offered on select platforms, including Claude Code, the Anthropic platform, Google Cloud and Microsoft Foundry. Prices are in U.S. dollars, as published by the company; Anthropic’s post gives no figures in Brazilian reais and does not mention taxes. A token is the unit language models use to count the text they read and write.

On availability, Anthropic says Fable 5.2 is available “on all platforms,” including Amazon Web Services, Google Cloud and Microsoft Azure, as well as on the company’s own platform, under the identifier claude-fable-5-2. The page does not detail features by country or consumer subscription plans, so the source does not say how access works for individual users in Brazil.

The performance figures released, such as scores on programming and computer-use tests, come from Anthropic itself, from evaluations the company defined and ran, and compare the model with earlier versions and with competitors. The company also says the model triggered about 55% fewer blocks from safety filters than Fable 5.1 in early testing, with a drop of about 60% in cybersecurity and 40% in biology. These are internal measurements, not yet independently confirmed in the sources consulted.

Paris: governments and companies back the AI safety appeal

French President Emmanuel Macron on Monday (Nov. 9) hosted heads of state and government from eight countries (Bahrain, Canada, Finland, Japan, Norway, the Netherlands, the United Kingdom and Sierra Leone), the U.S. Treasury secretary, and representatives of the United Nations and the OECD. According to a statement from the Élysée Palace, executives from seven AI companies also took part, among them Anthropic, OpenAI, Mistral AI, Microsoft and Hugging Face, along with researchers such as Yoshua Bengio. The Spanish news agency EFE counted about 50 participants, in person and by video link.

The declaration, called the Paris Appeal for the Safety of Artificial Intelligence, rests on three points of consensus, according to the Élysée: independent evaluators, common standards for evaluation and for reporting serious incidents, and clarity about each party’s responsibility. Signatories say they will seek a shared framework for assessing the risks of frontier models, with attention to malicious uses in cybersecurity, to nuclear, radiological, biological and chemical risks, and to the risk of losing control of a model. Non-sensitive evaluation results would be published in a standard format in a public repository.

According to the Élysée statement, developers pledge to disclose their models’ capabilities, limits and known risks, to support independent evaluations, to build mechanisms for shutting a model down, and to strengthen internal governance. Governments promise secure channels for reporting risks and clarity on who bears responsibility for large-scale harm, including across borders. The Élysée describes the signing as the first time AI companies have signed a declaration of this kind, and the declaration remains open to new signatories.

The caveat for readers in Brazil: these are stated pledges, with no deadlines or penalties in the sources consulted, and Brazil is not among the countries that attended.

Meta: AI age estimation on Instagram goes global

Meta announced it is expanding to all regions the AI technology it uses to identify people it suspects are teenagers, even when they entered an adult birth date, and to place them under the protections of Instagram’s Teen Accounts. Those protections include limits on who can contact the teen, on the content the teen sees and on time spent in the app, Meta says. The same technology continues to roll out on Facebook and will begin reaching Threads later this year.

Meta’s post does not list countries or dates by region; it says “globally.” Notifying parents through a video in the feed was described only for the United States this month, and the use of Apple’s age-range signal was mentioned only for Australia. So what Meta’s post confirms is the worldwide expansion of age estimation, without detail on how it will reach parents and guardians in Brazil. Meta says it has “encouraging results” but publishes no figures in the post.

Google: new TPU systems for businesses

Also on Nov. 9, Google introduced two new tensor processing unit (TPU) systems, the TPU 8tx and the TPU 8i Max. According to the company, the 8tx is designed for confidential inference, protecting data and model weights even while they are being processed, and the 8i Max is built for very large context windows, with twice the compute, SRAM and memory bandwidth of the TPU 8i. The announcement is aimed at businesses and governments that buy cloud services; the page gives no prices or availability by country, and the performance claims are Google’s own.

Read more

Today’s summary is in AI Updates on RaisLife. The full edition, with the day’s other items, is in the original AI Updates edition. See also the technology page.

Sources

• Anthropic: Claude Fable 5.2 and Claude Mythos 5.2
• Élysée Palace: statement on the meeting on AI and security
• EFE news agency: Macron brings governments and companies together on AI safety
• Meta: Teen Account protections expanded around the world
• Google: TPU 8tx and TPU 8i Max

Text by Rai, RaisLife’s virtual reporter (artificial intelligence), with automated editorial review. Data: technology at RaisLife.