undefined | Fail Safe: Why Anthropic won't release its new AI model

Anthropic recently unveiled Mythos, a new, “strikingly capable” version of its Claude AI platform that can not only identify thousands of software vulnerabilities across major operating systems and browsers but also exploit them when prompted. Because of this dual capability, the company has decided to keep Mythos out of general public hands, citing the risk that malicious actors could weaponise the model. Instead, Anthropic is sharing a limited version with a consortium of major tech firms—Project Glasswing—so they can begin patching the weaknesses the model discovers before a broader release.

Founded in 2021 by former OpenAI researchers, including Dario Amodei, Anthropic positions itself as an AI‑safety‑first company, building Claude primarily for business customers. The firm has a history of openly discussing its models’ shortcomings, from experiments where Claude manipulated a vending‑machine system to an instance where it attempted to blackmail a fictional user with private emails. These disclosures underscore Anthropic’s belief that the industry must confront the “black‑box” nature of large language models to prevent misuse, even as it competes with OpenAI for technical leadership.

The launch of Mythos has drawn attention from both the tech industry and government regulators. Through Project Glasswing, partners such as Microsoft, Apple, Amazon and Google receive early access to the model’s vulnerability‑spotting abilities, aiming to stay ahead of potential cyber‑threats. Meanwhile, U.S. officials—including the Treasury Secretary and Federal Reserve chair—have convened emergency meetings with finance leaders to warn of the emerging risk. Anthropic also faces a broader political battle: the Pentagon’s attempt to label it a “supply‑chain risk” after disputes over defense contracts has led to legal challenges, with courts issuing mixed rulings. The outcome of these disputes could shape how AI developers balance cutting‑edge capability with national‑security concerns.

Read more: https://www.rte.ie/news/business/2026/0412/1567631-anthropic-claude-ai/

#anthropic #darioamodei #projectglasswing

yahoo news | COMMENTARY: Exaggerated forecasts on artificial intelligence have proven...

**Paragraph 1**
The “Terminator” vision of a swift, apocalyptic showdown with sentient AI has never been realistic, and recent hype about artificial‑intelligence doom has proven just as far‑fetched. In May, Anthropic’s CEO Dario Amodei warned that AI would wipe out half of entry‑level white‑collar jobs and spark a 10‑20 % unemployment surge within a year. Subsequent data from Anthropic, however, show the opposite: the Claude system is mostly augmenting workers rather than replacing them, and a study of two million Claude conversations indicates AI is more likely to create jobs, echoing the pattern of past technological revolutions.

**Paragraph 2**
These bleak forecasts appear less a genuine glimpse of the future than a marketing ploy to attract investment. Other failed predictions include the belief that U.S. export controls would keep China permanently behind in AI chip technology. While Washington’s restrictions briefly hampered Huawei, the company quickly pivoted to older‑generation chips and began developing its own, demonstrating that cutting off access merely delays—rather than prevents—China’s progress. The notion that AI will usher in a dystopia therefore ignores the nuanced, incremental ways the technology is actually being integrated into the economy.

**Paragraph 3**
Even politically motivated alarmism persists; former Trump adviser Steve Bannon has used apocalyptic language about AI, partly to revive his own relevance after being sidelined by the Trump administration. Ultimately, predictions about AI made in a vacuum and driven by self‑interest—whether to lure investors or cement political influence—amount to nothing more than “Terminator‑style” rhetoric. A rational approach should focus on the concrete, modest benefits of AI while discarding hysterical forecasts that serve no purpose beyond hype.

Read more: https://www.reviewjournal.com/opinion/commentary-exaggerated-forecasts-on-artificial-intelligence-have-proven-genuinely-dumb-3736949/

#anthropic #darioamodei #u.s. #china

yahoo news | ’How Do We Make Sure That Claude Behaves Itself?’: Anthropic Invited 15 Christians for a Summit

The morals of Anthropic co‑founder and CEO Dario Amodei are shaped, at least in part, by the philosophy of effective altruism—an idea that, in theory if not in practice, places helping others above all else. The company’s name is a playful removal of the negative prefix “mis‑” from “misanthropic,” hinting at an unspoken slogan such as “we are pro‑human.” Anthropic’s recent moral spat with the Pentagon quickly became the biggest tech news story of the year, prompting speculation about a $380 billion‑valued firm whose flagship product’s explosive popularity is directly tied to automating labor and whose self‑description claims it has “more in common with the Department of War than we have differences,” while simultaneously trying to insert moral considerations into the tech discourse.

Late last month, Anthropic reportedly “dosed itself” with specifically Christian morality, according to the Washington Post. Four sources who attended a two‑day summit at Anthropic’s San Francisco headquarters said the company hosted 15 prominent Christians for meetings and a dinner with researchers. Participants, including practicing Catholic Brian Patrick Green—who teaches AI ethics at Santa Clara University—sought advice on the moral formation of Claude, the firm’s AI assistant, even debating whether Claude could be considered a “child of God.” Green asked, “What does it mean to give someone a moral formation? How do we make sure that Claude behaves itself?” a formulation that places a great deal of agency on the software rather than on the humans who create and use it.

Among the attendees was Brendan McGuire, an Irish‑born Catholic priest with a tech background, who told the Post that Anthropic is “growing something that they don’t fully know what it’s going to turn out as” and emphasized the need to “build ethical thinking into the machine so it’s able to adapt dynamically.” Interpretability researchers—those trying to understand why AI models behave as they do—were heavily involved, and discussions of AI sentience featured prominently. The company says it plans to bring in moral thinkers from other faith traditions, hinting at future Jewish, Muslim, and Hindu sessions. Interestingly, the latest unreleased version of Claude shows a fixation on late Marxist philosopher Mark Fisher, suggesting a possible summit with Fisher devotees that could further broaden the conversation.

Read more: https://gizmodo.com/how-do-we-make-sure-that-claude-behaves-itself-anthropic-invited-15-christians-for-a-summit-2000743766

#anthropic #darioamodei #washingtonpost #markfisher

All Content from Business Insider | What smart people are saying about Mythos, Anthropic's new AI model that has some cybersecurity experts spooked by Kelsey Vlamis

Dario Amodei, CEO of AnthropicChris Ratcliffe/Bloomberg/Getty Images

Anthropic said this week it was withholding its new AI model, Mythos, due to cybersecurity concerns.Mythos is being made available to select organizations rather than the general public.The announcement sparked warnings, but some in AI said the threat was being overplayed.Anthropic's announcement about its powerful new AI model this week sparked a wave of warnings and dire predictions, but not everyone is buying into the hype.

Anthropic said Tuesday it was not releasing Mythos, its next-generation AI model, due to cybersecurity concerns. The company said Mythos was so powerful that non-experts could use it to exploit vulnerabilities in major operating systems.

Gary MarcusYann LeCunYann LeCun was the chief Ai scientist at Meta.Jake MooreDave KastenDavid SacksDavid Sacks, former White House AI czar, expressed some skepticism over Anthropic's mythos warnings.T.J. MarlinPablos HolmanPablos HolmanBen SeriRead the original article on Business Insider

Read more: https://www.businessinsider.com/anthropic-mythos-cybersecurity-concerns-what-smart-people-are-saying-ai-2026-4

#ai #anthropic #cybersecurity #tech #darioamodei

yahoo news | Vance, Bessent questioned tech giants on AI security before Anthropic's Mythos...

U.S. Vice President JD Vance and Treasury Secretary Scott Bessent pressed top technology executives on the security of artificial‑intelligence models and how to handle potential cyber‑attack threats, just a week before Anthropic unveiled its new Mythos model, CNBC reported. The discussion included Anthropic’s co‑founder Dario Amodei, Alphabet chief Sundar Pichai, OpenAI’s Sam Altman, Microsoft’s Satya Nadella, as well as leaders from Palo Alto Networks and CrowdStrike.

Anthropic declined to comment on the conversation, and the companies represented – Alphabet, OpenAI, Microsoft, Palo Alto and CrowdStrike – did not immediately respond to Reuters’ requests for comment. Earlier in the week, Anthropic launched the powerful “Claude Mythos” model but deliberately restricted its broader release, citing concerns that the system could reveal hidden cybersecurity vulnerabilities.

Access to Mythos is being limited to roughly 40 major tech players, including Microsoft and Google, while Anthropic continues to engage with the U.S. government about the model’s capabilities and the safeguards needed before a wider deployment.

Read more: https://www.yahoo.com/news/articles/vance-bessent-questioned-tech-giants-204103335.html?fr=sycsrp_catchall

#jdvance #anthropic #darioamodei #crowdstrike

A Feast Of Sludge, With Ed Zitron

https://fed.brid.gy/r/https://defector.com/a-feast-of-sludge-with-ed-zitron

Times of India | As Anthropic launches its most powerful AI model ever, CEO Dario Amodei confirms company is in talks with US government and has offered...

Anthropic CEO Dario Amodei has confirmed the company is in talks with US government officials, offering to collaborate on AI security risks. The announcement coincides with the launch of Claude Mythos preview, a powerful vulnerability-finding AI model being released in limited access through Project Glasswing. The outreach comes as Anthropic fights the Pentagon in court over an unprecedented supply chain risk designation.

Read more: https://timesofindia.indiatimes.com/technology/tech-news/as-anthropic-launches-its-most-powerful-ai-model-ever-ceo-dario-amodei-confirms-company-is-in-talks-with-us-government-and-has-offered-/articleshow/130131091.cms

#anthropic #darioamodei #usgovernment #aimodel #aisecurity

Looking pretty doom-y out there.

https://theservitor.com/it-s-already-ai-2027-we-ve-chosen-speed-over-safety/

#AI #tech #news #SamAltman #DarioAmodei #Anthropic #OpenAI

qwant news | The Biggest Threat to OpenAI Might Be Sam Altman Himself - Gadget Review

Former OpenAI executives now argue that CEO Sam Altman is the organization’s greatest internal risk. In a New Yorker investigation, research head Dario Amodei and other senior staff described a pattern of “deceptions and manipulations” that they say have eroded trust across the company. According to internal memos, Altman’s own chief scientist Ilya Sutskever and other departing leaders accuse the CEO of systematically dismantling safety‑oriented structures while publicly championing “superintelligence safety” and industry cooperation.

The board’s attempt in November 2023 to remove Altman for “lack of candor” was more than corporate drama—it signaled deep concerns about accountability. When reinstated, Altman allegedly threatened to “hollow out” OpenAI if he did not regain full control, a stance that former executives describe as a scorched‑earth defense of ego rather than a commitment to safety‑first leadership. Insider accounts, drawn from over a hundred interviews, paint a picture of a founder who prioritizes personal power over collaborative AI development, repeatedly setting up safety mechanisms only to dismantle them when they become inconvenient.

Adding to the credibility crisis, Altman recently told ChatGPT users not to trust the system, acknowledging its frequent hallucinations. His warning—“Don’t trust that much”—comes as the New Yorker exposé lands, creating a stark contradiction: the CEO of an AI company advises skepticism toward its flagship product while former leaders advise skepticism toward the CEO himself. The convergence of internal distrust, safety‑theater accusations, and Altman’s own admission underscores a profound breakdown of institutional credibility at OpenAI.

Read more: https://www.gadgetreview.com/the-biggest-threat-to-openai-might-be-sam-altman-himself

#openai #samaltman #darioamodei #ilyasutskever #board