Was hat die US-Regierung für Anthropic konkret beschlossen?
Dem Unternehmen zufolge hat die US-Regierung Exportkontrollen für Mythos und Fable 5 angeordnet, zwei seiner leistungsstärksten Modelle. Sie dürfen demnach nicht mehr von Ausländern verwendet werden. Es ist eine ungewöhnlich weitreichende Sperre. Sie trifft Anthropic zufolge auch Ausländer, die sich in den USA befinden, darunter auch ausländische Anthropic-Mitarbeiter. Das Unternehmen hat nach der Anordnung beschlossen, den Zugang zu den Modellen für alle Kunden zu blockieren, also auch für Amerikaner. Fable 5 und Mythos werden erst einmal komplett abgeschaltet.
Was macht die Modelle Fable 5 und Mythos so besonders?
Anthropic selbst hielt seine Künstliche Intelligenz Mythos im April für so gefährlich, dass es die Software nicht allgemein zugänglich machen wollte und sie nur einer Handvoll amerikanischer Unternehmen zu Testzwecken zur Verfügung stellte. Dazu gehörten unter anderem Amazon, Apple, Microsoft, Google , Nvidia, Cisco oder auch die Linux-Stiftung. Insbesondere ging es um Missbrauchsrisiken durch Hacker ob der Fähigkeiten des Modells, bislang unbekannte Sicherheitslücken in IT-Systemen aufzuspüren und auszunutzen. International diskutierten Regierungen über die Auswirkungen für die nationale Sicherheit. Mit Claude Fable 5 hat Anthropic vergangene Woche eine Art abgespecktes Mythos-Modell angekündigt. Dieses ist so programmiert, dass es Anfragen zu bestimmten sensiblen Themen abblockt und an Claude Opus 4.8 weiterleitet.
Das britische Institut für KI-Sicherheit (AISI) hat für eine unabhängige Untersuchung Zugriff auf Mythos erhalten und dabei festgestellt, dass die IT-Sicherheitsfähigkeiten des Modells tatsächlich deutlich über denen von Vorgängermodellen liegen. Beispielsweise konnte das Programm erstmals einen simulierten 32-Schritt-Angriff auf ein Unternehmensnetzwerk vollständig und autonom abschließen. Allerdings haben auch andere Anbieter Modelle mit ähnlichen Fähigkeiten entwickelt, etwa Google mit Big Sleep. Das Modell GPT-5.5 des ChatGPT-Entwicklers Open AI schnitt im Test der AISI sogar genauso gut ab wie Mythos.
Was hat die US-Regierung zu dem Verbot bewogen?
Das ist noch etwas unklar. Anthropic sagt, die Regierung habe Bedenken um die nationale Sicherheit geäußert, aber keine spezifischen Details genannt. Nach Erkenntnissen des Unternehmens habe die Regierung von einer Methode erfahren, die Sicherheitsmechanismen von Fable 5 mit einem sogenannten „Jailbreak“ zu umgehen. Das „Wall Street Journal“ schrieb, diese Schwachstelle sei von Forschern des Technologiekonzerns Amazon gefunden worden. Sie hätten Fable 5 dazu gebracht, Informationen zu liefern, die für Cyberattacken nützlich sein könnten. Amazon-Vorstandschef Andy Jassy habe ranghohe Vertreter der Regierung wie den Finanzminister Scott Bessent auf die Schwachstelle hingewiesen. Die Onlinepublikation „Semafor“ brachte die Sperre mit Mutmaßungen im Weißen Haus in Verbindung, wonach Gruppen mit Verbindungen nach China sich Zugang zu Mythos verschafft hätten.
Warum sind keine anderen Anbieter wie Open AI oder Google betroffen?
Das fragt sich Anthropic nach eigener Darstellung selbst. Das Unternehmen beschreibt die vermeintliche Schwachstelle als nicht gravierend und sagt, sie lasse sich auch in KI-Modellen anderer Anbieter finden, etwa GPT-5.5 von Open AI. Anthropic hat freilich ein besonders kompliziertes Verhältnis zur US-Regierung. Das Unternehmen liefert sich seit Monaten einen Streit mit dem Verteidigungsministerium um die Frage, zu welchen Zwecken die Regierung KI-Modelle nutzen darf. Das Pentagon erklärte Anthropic Ende Februar zum Risiko für die nationale Sicherheit, woraufhin das Unternehmen die Regierung verklagt hat.
Was bedeutet die Entscheidung für das Geschäft Anthropics?
Da Mythos und Fable noch einen überschaubaren Nutzerkreis haben, dürften sich die unmittelbaren Auswirkungen auf das Geschäft in Grenzen halten. Aber die Sperre trifft das Unternehmen in einer Zeit, in der es seinen Börsengang vorbereitet und sich als wachstumsstark präsentieren will. Anthropic bezifferte seinen aufs Jahr hochgerechneten Umsatz kürzlich auf 47 Milliarden Dollar. Im April war noch von 30 Milliarden Dollar die Rede, im Februar von 14 Milliarden Dollar.
Welche Auswirkungen hat die Sperre für europäische Unternehmen?
Die kurzfristigen Auswirkungen dürften sich in Grenzen halten. Mit der IT-Sicherheit oder Biologie verbundene Aufgaben oder Anfragen an Fable 5 filterte Anthropic schon zuvor heraus und leitete sie auf das nächstbessere Modell Opus 4.8 um. Und Mythos war ohnehin nur einigen wenigen Unternehmen zugänglich. Ein Großteil der Welt konnte die besonderen Stärken der Modelle also schon vor dem Exportstopp nicht nutzen.
Hinzu kommt, dass Unternehmen für eine optimale Nutzung der Modelle im IT-Sicherheitskontext qualitativ hochwertige Kontextinformationen zum Quellcode ihrer Software oder zu ihrer Infrastruktur benötigen, sagt Fachmann Mirko Ross, Geschäftsführer des IT-…
Read the full article at Frankfurter Allgemeine (FAZ) →📄Source document: Anthropic→25 reports
heise onlineIndependentCenter2 days agoDespite the U.S. government's order to restrict access to Anthropic's AI models Fable and Mythos for foreign entities, several companies reportedly still have access to preview versions through the 'Project Glasswing' initiative. The Mythos Preview was provided to a range of companies, primarily in the IT and finance sectors. Anthropic has avoided general release of Mythos, citing safety concerns. A version called Fable 5 was briefly available but was taken offline following the government directive. Companies such as Dragos and Cisco confirmed they still have access to older versions of Mytho
Bias read (Center): The article presents factual information without overtly favoring any side. It reports on corporate access to AI models despite regulatory restrictions, quoting multiple companies and referencing Anthropic's statements. There is no evident bias in language, sourcing, or framing.
Official sources cited
- press release Bloomberg
- organisation Anthropic
Deutsche Welle (English)State / PublicCenter2 days agoThe U.S. government has imposed strict export controls on Anthropic, an AI company based in San Francisco, restricting access to its advanced AI models, including Claude Fable 5 and Mythos 5. The move was justified by national security concerns related to 'jailbreaking' techniques that could bypass AI safety protocols. Anthropic stated these issues were overstated and present in competing AI systems. As a result, the company suspended global access to its models. This decision followed Anthropic's announcement of plans for a public listing later this year. Tech industry leaders expressed alarm
Bias read (Center): The article presents facts without overtly favoring any side. It reports on the U.S. government's actions, Anthropic's response, and reactions from the tech community without using biased language or selectively omitting perspectives.
Official sources cited
- government U.S. Commerce Department
- organisation Anthropic
Tagesschau (ARD)State / PublicCenter3 days agoThe U.S. government has imposed restrictions on Anthropic's AI models, causing global uncertainty within the AI industry. At a technology conference in San Francisco, Databricks CEO Ali Ghodsi emphasized the importance of open-source AI models as an alternative, warning against potential restrictions on them. Anthropic released a less powerful version of their model called 'Claude Fable 5,' which lacks capabilities in areas like biochemistry and cybersecurity. The more advanced 'Mythos 5' model was reportedly restricted due to security concerns.
Bias read (Center): The article presents facts without overtly favoring any side. It reports on the U.S. government's actions, quotes a business leader expressing concern over regulation, and provides technical details about the AI models involved. There is no clear ideological framing or biased language.
taz – die tageszeitungIndependentCenter5 days agoThe article discusses Anthropic, a U.S.-based AI company, which recently removed two of its latest AI models, Mythos 5 and Fable 5, from public access following a government order restricting their availability to non-Americans. The piece reflects on a conversation the author had with two individuals knowledgeable in computer science a year earlier, during which they discussed Anthropic’s new AI model, Claude. At that time, Anthropic had raised concerns about the risks posed by AI technology and called for greater regulation of the industry. The article questions who controls such powerful AI,
Bias read (Center): The article presents a balanced discussion of Anthropic's actions, highlighting both the company's warnings about AI risks and the regulatory response from the U.S. government. It does not exhibit strong ideological framing, loaded language, or one-sided sourcing.
Frankfurter Allgemeine (FAZ)Independent🔒Center5 days agoThe U.S. government has imposed export controls on two advanced AI models developed by Anthropic—Mythos and Fable 5—preventing their use by non-Americans, including foreign employees based in the U.S. Anthropic has since blocked access to these models for all customers, including Americans. The models were considered highly capable and potentially dangerous due to their ability to detect and exploit unknown security vulnerabilities in IT systems. The decision has raised concerns about national security implications globally.
Bias read (Center): The article presents factual information without overtly favoring any political perspective. It reports on the U.S. government’s actions regarding AI export controls and includes details provided by Anthropic and other entities involved. There is no evident ideological framing or biased language.
Frankfurter Allgemeine (FAZ)Independent🔒Center5 days ago Anthropic vs. U.S. Government: The AI Warning ShotThe U.S. government has imposed an export ban on two advanced AI models developed by Anthropic—Mythos 5 and Fable 5—preventing access for all foreign entities, including allies. This decision has forced Anthropic to shut down these models entirely, even for American users. The company argues that the restrictions are disproportionate, citing minimal vulnerabilities in the models and noting similar features in competitors' offerings. The U.S. government reportedly based the decision on security concerns, though Anthropic claims the measures could hinder domestic applications of the technology.
Bias read (Center): The article presents both the U.S. government's stance and Anthropic's response without overtly favoring either side. It includes direct quotes from Anthropic and references the stated rationale behind the export ban, maintaining a balanced tone.
Official sources cited
- government U.S. Government Export Ban on Anthropic Models
- organisation Anthropic Statement on Export Restrictions
Der SpiegelIndependentLeft6 days agoThe article discusses an export ban on AI models imposed by the U.S. Department of Commerce, which was triggered by a dispute between the AI company Anthropic and the Trump administration. The ban initially targeted foreign users due to security concerns but was expanded to all users after Anthropic found it too burdensome to comply with the citizenship check requirements. The article highlights the broader implications of this action, emphasizing how the U.S. is using AI control as a matter of national power and how Trump's policies might undermine America's position as an innovation leader.
Bias read (Left): The article frames the U.S. actions under the Trump administration as overly aggressive and potentially harmful to American innovation leadership, suggesting a critical view of Trump's approach. It emphasizes Europe's vulnerability and calls for greater digital sovereignty, aligning with progressive
Official sources cited
- government U.S. Department of Commerce
- organisation Anthropic
taz – die tageszeitungIndependentCenter6 days ago Blocking of Anthropic's AI models: A warning with good timingThe U.S. government has ordered the California-based AI company Anthropic to restrict access to its powerful models 'Fable 5' and 'Mythos 5' for foreign users and employees due to national security concerns. Anthropic argued it could not differentiate between domestic and international users and therefore blocked access for everyone. While this decision does not immediately affect Europe, it highlights the speed at which the U.S. government is willing to restrict software access for foreign actors. The two models are considered particularly effective in identifying software vulnerabilities and
Bias read (Center): The article presents the situation factually without overtly favoring any side. It mentions the U.S. government's actions and Anthropic's response neutrally, while acknowledging uncertainties regarding the capabilities of the AI models.
Süddeutsche ZeitungIndependent🔒Center6 days agoThe article discusses the tension between former U.S. President Donald Trump and Anthropic, an artificial intelligence research company. It raises questions about whether creators of AI have lost control over their technology.
Bias read (Center): The article appears to present a balanced discussion without overtly favoring either side. It frames the issue as a general question about AI control rather than taking a stance on Trump or Anthropic specifically.
heise onlineIndependentCenter6 days ago Restriction of Anthropic's AI models: China is said to have had access to MythosThe U.S. government has ordered the shutdown of Anthropic's AI models Fable 5 and Mythos 5 due to concerns that a group linked to the Chinese government may have accessed Mythos 5. This information comes from the U.S. magazine Semafor, citing an anonymous source. The restriction of access to Mythos 5 was supposed to be strictly controlled. If China had access, it could pose a risk to U.S. national security. However, this has not been confirmed yet. Previously, there were reports that security measures of Fable 5 could be bypassed, which led to the ban. Anthropic released Fable 5 last Tuesday,
Bias read (Center): The article presents facts without overtly favoring any side. It cites an unnamed source and mentions potential risks but does not take a clear stance on the allegations or their validity.
Tagesschau (ARD)State / PublicCenter8 days ago US government order: Anthropic blocks access to AI modelsThe article reports that Anthropic has restricted access to its AI models following an order from the U.S. government.
Bias read (Center): The article presents a factual report without evident framing or slant. It does not include commentary, opinion, or biased language. The focus is on the action taken by Anthropic in response to a government directive, with no indication of favoring one side over another.
heise onlineIndependentCenter8 days ago Comment: The best advertising for alternative AI manufacturersThe article discusses the global ban on AI models Fable and Mythos due to actions by the U.S. government. It notes the limited direct damage but emphasizes the strong signal sent by this move.
Bias read (Center): The article presents a neutral observation about the impact of the U.S. government's actions without overtly favoring any side. The tone is analytical rather than opinionated, focusing on the implications of the ban rather than taking a stance on it.
Tagesschau (ARD)State / PublicCenter8 days ago Security concerns: Anthropic must lock latest AI versionsAnthropic, a US-based AI developer, has blocked access to its latest AI models 'Fable 5' and 'Mythos 5' worldwide following an order from the US government citing national security concerns. The models, known for their ability to detect long-hidden software vulnerabilities, were previously used by US agencies and select companies for security purposes. Anthropic stated that the government did not provide further details on its security concerns.
Bias read (Center): The article presents factual information without overtly biased language or framing. It reports on the government's directive and Anthropic's response without taking a stance or emphasizing one perspective over another. The context of national security is mentioned neutrally, and the article does not editorialize or omit relevant information.
Official sources cited
- organisation Anthropic
- government U.S. Department of Commerce
Deutsche Welle (English)State / PublicCenter8 days ago Anthropic cuts top-tier AI access after US foreigner banAnthropic has restricted access to its advanced AI models, Fable 5 and Mythos 5, for foreign nationals due to national security concerns.
Bias read (Center): The article presents a factual statement without editorializing, loaded language, or one-sided sourcing. It reports on Anthropic's action as stated by the company, without implying approval, criticism, or bias.
Frankfurter Allgemeine (FAZ)Independent🔒Center8 days ago Dispute with Trump administration: Anthropic shut down AI models after ban for foreignersAnthropic has withheld two new AI models due to disagreements with the U.S. government over security concerns.
Bias read (Center): The article presents a factual statement without overtly biased language or emphasis. It does not take a stance on the dispute between Anthropic and the Trump administration, merely stating that Anthropic withheld models due to disagreements over security concerns.
Der SpiegelIndependentCenter8 days ago Anthropic: AI model Mythos taken off the grid following US export restrictionsAnthropic has deactivated its AI models 'Mythos 5' and 'Fable 5' following security concerns raised by the U.S. government. Foreign nationals are no longer allowed access to these models due to security reasons.
Bias read (Center): The article presents factual information without overtly biased language or emphasis. It reports on the deactivation of AI models based on U.S. government security concerns without taking a stance or providing additional context that would indicate a particular ideological leaning.
Official sources cited
- government U.S. Government Security Concerns
heise onlineIndependentCenter11 days ago Trump administration reportedly misses a civilian AI test siteThe White House appears to be tightening control over particularly powerful AI models. The 'Center for AI Standards and Innovation' (CAISI), responsible for model testing and security assessments, has reportedly been instructed by government officials to temporarily halt publishing model evaluations until the implementation of President Trump's new AI security decree. This decree includes provisions allowing U.S. agencies voluntary access to especially powerful AI models for up to 30 days and establishing a classified benchmarking process with security-focused agencies playing a central role.据
Bias read (Center): The article presents facts without overtly biased language or selective sourcing. It reports on actions taken by the Trump administration and quotes both internal sources and official statements, providing a balanced view of the situation.
HandelsblattParty-aligned🔒Center11 days agoAnthropic has released a new high-performance AI model called 'Fable'.
Bias read (Center): The article reports on the release of a new AI model without taking a stance or showing bias in language, sourcing, or emphasis.
Der SpiegelIndependentCenter11 days ago Anthropic: Controversial AI model Mythos only available to a limited extentAnthropic has publicly released an AI model called Mythos in a limited form. Despite restrictions, concerns remain about potential military applications and use for cyberattacks.
Bias read (Center): The article emphasizes concerns about military use and cyberattack risks without mentioning Anthropic's safety measures or providing balanced context, creating a cautionary rather than neutral framing.
heise onlineIndependentCenter11 days ago This is going to be expensive: Anthropic's Claude Mythos 5 is being released as Fable 5 with restrictions.Anthropic's Claude Mythos 5 model is available to the NSA and select partners, while a restricted public version called Claude Fable 5 is being released without a subscription model.
Bias read (Center): Headline frames exclusive model access and public restrictions with cost-critical language ('Das wird teuer'), suggesting editorial skepticism about access disparity and value, though the body is factual.
Germany