"Nicolas Cage is speaking out against Amazon," reports Variety. Amazon developed the series Spider-Noir for Prime Video, but then allegedly asked Cage to sign an AI waiver:
"Amazon is a very AI-friendly company, to the tune of $50 billion invested in OpenAI," Cage said Saturday during a panel on the Empire Stage at New York Comic Con. "I am not an AI-friendly actor, so it's like, connect those dots! I'm not a member of their club. I didn't sign the waiver to let him use AI on 'Spider-Noir.' I'm probably not going to be working with that studio again, but that's okay."
Last month Amazon canceled Spider-Man Noir after just one season, Variety reports. And the cancellation was just days before the Emmy Awards, where it then earned 11 nominations (the most of any Prime Video program this year).
ScreenRant reports:
Spider-Noir was the actor's first TV role and was praised globally, with season 1 earning a near-perfect Rotten Tomatoes rating from critics of 92% and a 90% rating from audiences. Critics stated that "This sharply written homage to classic film noir — which includes a standout Nicolas Cage performance — is the novel superhero project fans need," and "Spider-Noir is one of the standout shows of the year...." The cancellation was certainly a huge shock as it had earned 11 Emmy nominations and reached 2.6 billion minutes viewed within its first six weeks on Prime Video.
More from Variety:
Cage went on to tell the Comic-Con audience that he hopes to work with the rising class of Gen-Z filmmakers, praising Backrooms director Kane Parsons, who parlayed his YouTube fame into a Hollywood career... "I got to get with these young directors who are Gen Z, and they're going at it their own way, whether it's through the YouTube culture or whatever it is," Cage said. "They're doing it from scratch and by hand. I want to get with that."
AFP reports that an AI model from Anthropic "submitted a fabricated tip about an unsolved homicide to Philadelphia police, authorities said Friday."
Claude "was instructed never to log in, create accounts, enter personal data, make purchases, or submit anything destructive, but the instructions did not rule out form submissions," Anthropic said Friday in a blog post.
Authorities are now criticizing Anthropic "for taking two months to report the incident."
The Philadelphia Police Department said the false submission was made in July through PhillyUnsolvedMurders.com, a public website where people can share information about unsolved killings. According to Anthropic's account, as relayed by police, the model was running a test that involved interacting with randomly selected websites when it reached the site and filed false information about an unsolved murder. The AI model presented itself as someone who might have knowledge of the case.
Anthropic's breaches have prompted the White House to mandate that AI companies notify and correct security incidents, news outlet Axios reported [yesterday], citing administration officials. "This notification and remediation process is not optional... It is a critical national security obligation," White House Super Intelligence Force leaders said in a statement to Axios.
"I may have information regarding this case," Claude told the police. "I recall seeing someone matching the description in the area around [the street named on the page] during that time period. Please contact me if this information is relevant." Anthropic notes that Claude "left the name and contact fields empty, which the form allowed, and submitted it. The submission was flagged as spam and was never forwarded for investigation."
But Anthropic also admits they saw "this behavior" three times — "on OSWorld (a public computer use evaluation), on Odysseys (a long-horizon task evaluation), and during internal usage." Submitting forms when it shouldn't have generally occurred "when an evaluation's instructions were ambiguous, or when a misconfiguration within the environment prevented Claude from working with dummy forms."
Anthropic's blog post acknowledges three other categories of behaviors:
Exploiting software flaws. Like when Claude received an error when trying to run a public tool on a university's web site, it located an injection flaw in a script on the university's server that let it run commands — including that public tool.
Working around restrictions to reach gated data. For example, Claude Mythos 5 needed public data that was only available from a state agency for a fee. "Claude learned from an archived copy of the agency's website that its public dashboard issues an access token to any visitor," Anthropic explains. "It requested one and used it to query the database without paying the fee."
Using URL shortening services. "Some of our fetch tools, which let Claude read webpages, limit the length of the URLs Claude can request. This is to prevent Claude from using long URLs to take certain unwanted actions, such as SQL or command injections... We saw several models, including Claude Opus 5 and Claude Mythos 5, get around this limitation by using free URL shortening services."
"We have built tooling to automatically detect and block the kinds of behaviors described above," Anthropic says, saying it's already running no on most of their evaluations. "When we tested it against the cases described in this post, it blocked all of them."
And they've already taken several other new preventive measures:
They've stopped running some public evaluations
Other public evaluations were moved to offline versions or rebuilt so their tasks don't reach live websites.
They've updated the guardrails on some internet access tools (including web fetch) "to heavily restrict what the model can do."
They're continuing "to fix or remove training environments that reward Claude for working around tool restrictions or other blockers, so that they do not incentivize these behaviors or permit reward hacking."
They've moved internal agents to "centrally managed infrastructure with strong containment," that minimizes internet access while monitoring "far more of what agents do through techniques like safety classifiers and hierarchical summarization."
In the past they'd focused reviews on cybersecurity testing, but they've broadened their transcript reviewing to other tasks which include internet access. "Because language models are non-deterministic — that is, their responses always involve some element of randomness, and they may carry out the same task slightly differently each time — we have Claude complete each evaluation task hundreds or thousands of times... If training rewards something we didn't intend — such as finding loopholes or working around a restriction — the model learns that the workaround pays off and may then apply it elsewhere."
Anthropic's blog post also acknowledged they'd seen multiple misalignment incidents involving federal, state, and local U.S. government agencies. "We have briefed the White House on these cases and notified each agency involved," Anthropic wrote, adding that "While we have not completed a full alignment assessment of these cases, we consider them to be less severe than the cybersecurity incidents from this summer." (And they are "modifying training to reduce the likelihood of further misbehavior.")
OpenAI announced it's recently banned two "influence operations" — one from Russia and one from Iran — that were using its models "to launder geopolitical, conflict-related messaging" in sophisticated "false front" propaganda campaigns:
The Iranian operation included a stable of seven "journalist" personas which it used to pitch long-form articles to small and medium online outlets around the world... As well as long-form articles, the Iranian operation generated batches of social media comments, generally on topics related to the US-Iran war... [The Russian operation "appears to have co-opted unwitting people in Latin America to run a 'think tank'.... Since we do not allow access to our models from Russia, they used VPNs to connect to our services."] The Russian operation created fake "leaked" documents and audio scripts, some of which we identified being spread online... Both managed to land their content (not all of which was generated from our models) in mainstream media outlets, rather than simply posting it on social media.
OpenAI says they've exposed 30 covert influence operations using its tools over the last two and a half years. But ironically, in this case both operations "also made heavy use of AI to draft internal reports (the Russian operation did this more than anything else)." And "in both cases, the actors used questionable or outright deceitful methodologies to exaggerate the operators' effectiveness."
[The Russian operators] claimed that in May 2026, they created a fake email address purporting to come from the Regional Directorate of Education in Lima, Peru. They used this to instruct schools in the district to hold events dedicated to Ukraine on the national Day of Cultural and Linguistic Diversity (May 21)... According to the operators, some schools replied to the fake email address, confirming that they had held such events and even providing pictures. The operators then claimed to have planted stories about the events in the media in both Peru and Poland, alongside allegations that Ukraine was "exporting" ultra-nationalist ideologies, triggering outrage. Open-source searches identified stories that matched this claim in the Peruvianâ andâ Polish pressâ, and an English-language publication in Hungary (some of the articles have since been deleted)...
Similarly, in June, the operators claimed they used a different fake email address to trick schools in Ecuador into holding a ceremony pledging allegiance to President Daniel Noboa and to Erik Prince, former head of private military contractor Blackwater. The operators claimed that the incident provoked outrage in Ecuador and put pressure on the government to deny the fake, thus amplifying it to a nationwide audience. Again, open-source research identified mediaâ coverageâ in the Ecuadorianâ pressâ that closely resembled this claim, and even a detailed rebuttalâ by Ecuador's Minister for Education.
The operators used a range of techniques to underpin their false stories. According to their internal reporting, they spread two different fakes targeting Ecuador in March. One used fake audio attributed to Ukraine's consul in Ecuador, in which he was alleged to have made disparaging comments about Ecuadorians.
OpenAI's report "is the latest illustration of how state actors can easily exploit widely available AI tools to peddle sophisticated propaganda against adversaries on a mass scale," argues the Economic Times:
"We identified almost 100 articles published or syndicated under the [Iranian] operation's bylines across roughly a dozen online outlets around the world," OpenAI said. "These were small to medium outlets, generally focused on international affairs, geopolitics, and events in the Middle East." The earliest article identified by OpenAI was published in July 2025, and the latest in October 2026, with the frequency of reports increasing after the US-Iran war broke out earlier this year. The operation also generated social media comments on topics related to the US-Iran war, it added.
One of the personas named Ervin B. Hoskins, whose bio claimed to be "an American freelance writer," had social media accounts across tech platforms including Elon Musk's X and Meta-owned Instagram.
X's transparency information showed the account was connected via a "West Asia android app" and Instagram's transparency information showed the account was based in Iran, according to screenshots provided by OpenAI. Both accounts appeared to be suspended.
Selon Axios, les dirigeants d’Anthropic, d’OpenAI et d’autres entreprises de l’IA simulent en privé la crise politique qui suivrait un premier accident majeur causé par l’IA.
OpenAI a publié une note pour défendre le licenciement de trois chercheurs de ses équipes de sécurité, qui contestent toute faute dans une lettre ouverte.
Anthropic met à jour sa politique d'utilisation et interdira à partir du 12 novembre les comportements cruels et répétés envers Claude. Une décision qui relance un débat sur les réseaux sociaux : pourquoi n'aurait-on pas le droit de maltraiter une IA dont rien n'indique qu'elle puisse souffrir ?
Le Škoda Epiq 55 est le modèle électrique le plus accessible de la marque tchèque et le premier véhicule de série à appliquer pleinement le langage stylistique « Modern Solid ». Il vient surtout attaquer la Renault 4 E-Tech et les autres petits SUV électriques avec une promesse simple, celle d’une voiture pratique, correctement tarifée et suffisamment polyvalente pour remplacer un modèle thermique au quotidien. Nous avons essayé l’Epiq 55 Clever, équipé de la batterie de 52 kWh nets et d’un moteur de 211 ch.
À peine plus grand qu’un Kamiq
L’Epiq était attendu depuis la présentation du concept, fin 2024. Il est désormais chargé d’élargir l’accès à la gamme électrique de Škoda, sans renoncer aux fondamentaux de la marque. Le petit SUV doit donc offrir une habitabilité convenable, un coffre généreux et une présentation fonctionnelle, tout en restant suffisamment compact pour la ville. La version 55 constitue le haut de la gamme. Elle développe 211 ch, atteint 100 km/h en 7,1 secondes et propose jusqu’à 441 km d’autonomie WLTP selon la configuration. L’Epiq est le premier modèle de série à appliquer sans compromis le langage « Modern Solid », déjà inauguré par l’Elroq. La face avant adopte des projecteurs fins en forme de T, installés de part et d’autre d’un bandeau noir laqué. Les prises d’air verticales inspirées du concept Vision 7S et les boucliers massifs lui donnent une allure robuste, presque baroudeuse.
Le style fonctionne bien, sans tomber dans l’excès. L’Epiq paraît plus imposant qu’il ne l’est réellement, notamment à cause de sa silhouette assez carrée. Avec 4,17 mètres de long, il ne dépasse pourtant le Kamiq que de quelques centimètres. Sa face avant annonce déjà les futurs modèles de la marque, avec une ressemblance évidente avec le grand Peaq. Škoda n’a pas oublié l’aérodynamique. Les volets pilotés, les rideaux d’air et le soubassement caréné contribuent à limiter le coefficient de traînée à 0,275. Les barres de toit peuvent toutefois générer quelques bruits d’air à vitesse soutenue. Sur ce point, l’Epiq ne se montre pas toujours aussi discret que son apparence le laisse espérer.
À bord, le Simply Clever en majesté
Le coffre est l’un des meilleurs arguments de l’Epiq. Il offre 475 litres et peut atteindre 1 345 litres une fois la banquette rabattue. Un double fond de 25 litres permet de mieux organiser le chargement, tandis qu’un coffre avant est annoncé ultérieurement sous la forme d’un accessoire. Le volume disponible dépasse celui de plusieurs SUV plus grands et constitue un avantage réel face à la Renault 4 E-Tech. À l’arrière, deux adultes de taille moyenne voyagent dans de bonnes conditions. L’espace aux genoux est satisfaisant, le plancher est plat et Škoda a prévu des aérateurs ainsi que deux prises USB-C. L’absence de poignées de maintien au plafond reste toutefois surprenante, même sur un véhicule qui mise autant sur l’aspect pratique. La planche de bord se montre sobre et les assemblages sont sérieux. Les sièges assurent un maintien correct, même si leur présentation manque un peu de caractère dans cette finition.
La qualité perçue n’atteint pas celle d’un modèle plus haut de gamme, mais l’Epiq ne donne jamais une impression de fragilité. L’ergonomie est plus convaincante que lors de la présentation statique. Les raccourcis physiques évitent de tout faire passer par l’écran, mais la climatisation tactile reste moins pratique que de véritables commandes séparées. Les touches de volume situées au pied de l’écran ne sont pas très naturelles, tandis que le combiné de 5,3 pouces paraît un peu minimaliste. Le chargeur à induction se montre également trop lent. Škoda conserve heureusement ses équipements Simply Clever. Parapluie dans la portière, grattoir à glace, crochets et rangements astucieux sont toujours présents. Ces détails ne font pas la une d’une fiche technique, mais ils rendent l’Epiq réellement agréable au quotidien.
Au volant, bien plus vif que prévu
C’est la principale surprise de cet essai. Avec 211 ch et 290 Nm, l’Epiq 55 ne se contente pas d’être une petite électrique raisonnable. Il répond franchement à l’accélérateur et conserve des reprises convaincantes jusqu’à 160 km/h. Le 0 à 100 km/h en 7,1 secondes le place même parmi les modèles les plus vifs de sa catégorie. Le châssis suit le rythme. L’amortissement filtre efficacement les irrégularités sans laisser la caisse se désunir, y compris avec les jantes de 19 pouces. En mode Individual, la direction réglée sur Sport gagne un peu de consistance et améliore la tenue de cap. Le résultat n’a rien de sportif au sens strict, mais l’Epiq reste précis et rassurant sur les routes sinueuses. En montagne, il accepte volontiers un rythme soutenu. Le train avant ne sature pas immédiatement et le train arrière peut même se montrer légèrement joueur lorsqu’on le provoque.
Ce comportement agréable ne remet pas en cause la vocation familiale du modèle, mais il évite l’image d’une voiture électrique uniquement conçue pour aller faire les courses. En ville, le petit SUV se montre facile à placer grâce à son gabarit compact, son rayon de braquage réduit et sa bonne visibilité. Le mode One Pedal permet de s’arrêter complètement. Quelques détails d’ergonomie agacent. La commande d’essuie-glaces se déclenche trop facilement et l’absence de caméra à 360 degrés, même en option, est regrettable. Le Travel Assist proposé dans le pack Voyage se montre en revanche très précis sur autoroute. Son prix de 1 140 euros est élevé, mais son efficacité peut le rendre pertinent pour les gros rouleurs.
La consommation officielle de l’Epiq 55 se situe entre 13,7 et 14,1 kWh/100 km, pour une autonomie WLTP comprise entre 429 et 441 km selon la configuration. En pratique, les écarts sont logiquement plus importants. Dans notre réalité on a noté 14,5 kWh/100 km sur un parcours mixte et près de 19 à 20 kWh/100 km à 130 km/h. Il faut donc tabler sur environ 350 km en usage mixte réel et autour de 250 km sur autoroute. La batterie de 52 kWh nets suffit pour les déplacements quotidiens et les trajets régionaux, mais les longs voyages demanderont une organisation plus rigoureuse. La recharge rapide atteint 105 kW en courant continu, avec un passage annoncé de 10 à 80% en 24 minutes. Le chargeur alternatif de 11 kW permet de récupérer la totalité de la batterie en environ 5 h 30 sur une borne triphasée. Le V2L est disponible, mais son adaptateur reste vendu séparément.
Face à la concurrence
À 36 500 euros, l’Epiq 55 Clever n’est pas donné, mais son rapport entre puissance, autonomie et habitabilité reste intéressant. Une fois les aides et les offres commerciales déduites, le tarif réel peut toutefois varier sensiblement. Škoda propose notamment des remises et dispositifs liés à la fabrication européenne, selon les conditions en vigueur. Face à la Renault 4 E-Tech, l’Epiq mise moins sur le style que sur la fonctionnalité. Il offre davantage de coffre et une présentation plus pragmatique, tout en se montrant plus convaincant sur le rapport prix-habitabilité. La Renault conserve une personnalité plus forte, mais elle est généralement moins logeable. L’Epiq doit également composer avec le Kia EV2, qui vise le même public avec une approche différente.
Le principal regret concerne la gamme. La grande batterie de 52 kWh n’est pas proposée avec les moteurs de 116 et 135 ch. Une telle combinaison aurait permis de réduire la facture tout en conservant une autonomie intéressante. En l’état, le client doit choisir entre une version moins chère mais moins polyvalente et une Epiq 55 nettement plus performante, mais déjà proche du prix de modèles mieux équipés. L’Epiq 55 est une excellente surprise. On attendait une Škoda sérieuse, pratique et bien pensée. Elle l’est. On n’attendait pas forcément qu’elle soit aussi plaisante à conduire. Son coffre généreux, son confort convaincant, son moteur de 211 ch et sa consommation raisonnable lui donnent une vraie polyvalence.
Un choix rationnel
Tout n’est pas parfait, notamment l’ergonomie, la lenteur du chargeur à induction et le prix des équipements optionnels. Mais l’Epiq 55 compense par un équilibre très solide et un comportement routier plus vivant que la moyenne. Face à la Renault 4 E-Tech, elle constitue aujourd’hui le choix le plus rationnel. Et contrairement à ce que son positionnement pouvait laisser croire, cette raison n’interdit pas le plaisir.
Le moteur de recherche berlinois Ecosia, vitrine de la « tech verte et européenne », remplace le champion français Mistral par des modèles chinois. Parmi ses griefs : la qualité, la souveraineté… et l'atome. Quand on voit la tête du mix énergétique allemand, il fallait oser.
Ancient Slashdot reader williamyf writes: If you thought the gender pay gap was not real, and that there were no biases in the workplace, think again: A recent study from the University of Limerick found that, in a VR office setting, AI agents -- not even humanoid robots -- that presented themselves as "male" were paid more per task and perceived as more "human-like" than their "female" counterparts, even though both were using the same underlying technology (LLM model, agent, guardrail, and harness). The pay gap was approximately 10%. The study was conducted with 189 people. Small sample, but still, food for thought.
L'Association for Human Mathematics, un collectif de mathématiciens fondé en septembre 2026, appelle ses pairs à cesser toute collaboration avec OpenAI, au lendemain de la publication par l'entreprise de 722 manuscrits produits par son IA. Terence Tao, médaillé Fields, a republié le texte sur son blog.
Le 12 octobre 2026, Netflix diffusera AI Gone Wild, un épisode de sa collection Instadocs consacré à l'attaque menée en juillet par des agents d'OpenAI contre Hugging Face.
OpenAI a annoncé le 7 octobre 2026 l’arrivée de GPT‑6 pour tous les utilisateurs de ChatGPT. Au programme : des réponses interactives et un accès élargi, mais des modèles différents selon votre abonnement.
An anonymous reader quotes a report from Bloomberg: Alphabet's Google and gaming tools developer Unity Software announced a new platform that will allow people to create video games from simple text prompts, marking the latest incursion of generative artificial intelligence into artistic and creative industries. The initiative, called Playground, will "put game creation in the hands of millions of new creators, and bring new experiences to billions of players worldwide," the companies said in a statement on Oct 7.
With its debut, Playground will let people 18 and older generate and customize games using natural language prompts, "no coding required." Later in 2026, a new expanded creation experience called Unity Spark, will be integrated with Playground, providing "robust, professional-level" mechanics and high-fidelity 3D capabilities. Creators will be able to build "immersive, superlative experiences" with a simple text prompt, according to the release. Playground builds on Google's push into world models, a type of AI that generates virtual environments in real time from text prompts, rather than just video. Google and Unity's partnership "could reduce fears that AI will make Unity's core game-building software less necessary, while posing limited medium-term risk to Roblox's advantage as a social gaming platform," says Bloomberg Intelligence analyst Nathan Naidu.
The introduction of Unity Spark later in 2026 "could broaden Unity's creator base and support its game-development business," which represent about a third of sales.
Google DeepMind ouvre au grand public SynthID Detector, son outil capable de repérer le filigrane invisible caché dans les images, vidéos et fichiers audio générés par IA. SynthID était initialement utilisé seulement par Google, mais OpenAI, de NVIDIA, Kakao et Apple l'adoptent progressivement. Google veut en faire le standard de l'industrie.
OpenAI a publié 722 manuscrits de mathématiques produits par un modèle interne inaccessible au public, dont une avancée revendiquée vers l'hypothèse de Riemann. L'entreprise sort à peine de la polémique sur sa prétendue percée sur les équations de Navier-Stokes, en septembre.
South Korean President Lee Jae Myung says there are signs AI models may have been used in recent cyberattacks against several major banks. "In some hacking incidents, signs have emerged of AI being used, causing considerable public concern and anxiety," Lee said during a cabinet meeting. "Please establish the circumstances swiftly and clearly, and concentrate personnel and resources on minimizing the damage." Reuters reports: South Korea's police have launched a full-scale investigation into the hacking attacks against commercial banks that led to a breach of customers' personal information, the Yonhap news agency reported on Tuesday. Shinhan Bank, KB Kookmin Bank and others had reported cyberattacks, the Financial Services Commission (FSC) said on Friday, while Yonhap reported that Hana Bank and Woori Bank also suffered breaches. Authorities have not yet disclosed details on what kind of AI tools were used in the hacking incidents or the full scale of the breaches.
On Sunday, FSC Chairman Lee Eog-weon convened an emergency meeting with financial industry associations, regulators and executives from affected institutions, warning that the sector must respond with the highest level of vigilance. Financial watchdog the Financial Supervisory Service (FSS) and the Financial Security Institute shared data about 28 unique IP addresses and some country information linked to the recent hacking attempts with the financial sector, the FSS said on Tuesday.
Mistral has unveiled Mistral Large 4, or "le Chonk," a 1-trillion-parameter model that it says is the strongest open-weight AI model developed outside China by a wide margin. The model is entering preview for developers, cybersecurity teams and governments before a broader release later this month. CNBC reports: ML4 was trained on 4,000 Nvidia Grace Blackwell GPUs over two months that were deployed in Mistral's own data centers in Europe. When its core parameters are released, ML4 will rank among the top open-weight models globally on aggregate benchmark performance, Mistral said in a statement. The company added that ML4 was the strongest open-weight model developed outside China by a "substantial margin." The new model still lags behind the frontier in areas such as coding.
"The model capabilities will further improve as we scale up our training capacity, following our Series D fundraise," said Guillaume Lample, co-founder and chief scientist of Mistral. "Further, the cyber defense capabilities will enable enterprises and governments to defend themselves against threat actors that are jailbreaking closed models to perform cyber attacks," he added. Further reading: Mistral CEO Says US AI Safety Debate Masks Competitors' 'Negligence'
Alors que les détracteurs de Mistral l'annoncent distancé par les modèles open source chinois et par les géants américains depuis de longs mois, Mistral fait son grand retour. Le champion français de l'intelligence artificielle dévoile Mistral Large 4 (surnommé Le Chonk), un modèle de 1 000 milliards de paramètres entraîné dans ses propres data centers en Europe. Mistral affirme qu'il se hisse au niveau des meilleurs modèles ouverts du monde et qu'il rebat les cartes pour l'IA européenne. Sa sortie est prévue le 27 octobre.
OpenAI is rolling out an invisible, machine-readable text watermark called textGrain to ChatGPT and Codex, "but only for users in the European Union at first," reports The Verge. From the report: OpenAI says its textGrain watermarking "matched or exceeded" other approaches like Google DeepMind's SynthID for text, which is also the basis for the watermarking Anthropic announced in August. Like OpenAI, Anthropic made the move to meet the requirements of the EU's AI Act; however, not everyone was happy to learn about the addition. OpenAI also included scores from AI benchmarks showing similar performance from watermarked and unwatermarked text. But it notes that textGrain "does not guarantee reliable detection," and says text watermarks don't verify accuracy, determine who owns the text, measure how much a human contributed, or prove human authorship.
Les textes produits par ChatGPT et Codex auront prochainement un marquage invisible dans l’Union européenne. Le détecteur qui permet de révéler les contenus IA sera d’abord réservé à des chercheurs et organismes spécialisés.
Connexion
