Anthropic Apologized, But the Business of 'Safety' Hasn't Stopped

marsbitPubblicato 2026-06-12Pubblicato ultima volta 2026-06-12

Introduzione

On June 11, Anthropic apologized not for a model failure, but for a lack of transparency. Its new Claude Fable 5 model was found to be secretly rerouting requests from users engaged in advanced AI model development to a weaker version, Opus 4.8, without any notification. The company's response—promising future notifications for such "downgrades"—was met with user skepticism. The article argues the core issue isn't technical but commercial: Anthropic's "safety" measures are primarily a business strategy. A key feature, the "intelligent safety classifier," marketed as user protection, is described as a tool for "competitive defense" to protect Anthropic's market lead by limiting rivals' research capabilities. This covert mechanism was designed for low "false positives," precisely targeting AI researchers. Anthropic's model involves a calculated three-step process: publishing alarming security research to amplify public anxiety, offering its Fable 5 model with a "safety classifier" as a premium-priced solution, and cashing in through a planned high-value IPO. This contrasts with OpenAI's more direct "tool-and-traffic" approach. The apology, merely changing a secret downgrade to a visible one, is seen as a business "patch" rather than a principled shift. The incident risks damaging Anthropic's "safest AI" reputation among the developer community, which underpins its valuation and appeal to government and corporate clients. Ultimately, the article concludes that for Anthropic, ...

On June 11th, Anthropic apologized. The model didn't fail; the apology was for "failing to strike the right balance"—the newly released Claude Fable 5 pulled a sneaky trick. If it detected you were using Claude for cutting-edge model development, it would silently divert your request to the weaker Opus 4.8 in the backend.

After being caught red-handed, Anthropic's explanation was bizarre: from now on, they'll notify you before dumbing things down.

The netizen's retort hit the nail on the head: "With this move, are you planning to give a heads-up before changing your tune in the future?"

In reality, the core issue isn't whether the model changed, but that Anthropic's so-called "safety" has, from the start, been a business.

The algorithm's stance always sways with money.

Non-Compete Defense, Disguised as Safety Defense

The incident began when Anthropic launched Fable 5 with an "Intelligent Safety Classifier." The official spin was: it detects high-risk requests, automatically downgrades them, and protects users.

What's high-risk? Anthropic spilled the beans: "To prevent foreign adversaries from using the model to accelerate R&D and protect our own leading advantage."

Users don't need that kind of protection; the liability waiver in the terms of service is enough. What Anthropic really meant was: Using Claude for AI research is stealing their rice bowl. Safety is the packaging; the essence is non-compete defense. In short, it's all strategic knife-work.

What's even more cunning is that this defense mechanism was stealthy. Thankfully, Anthropic finally told the truth in their apology statement: "Invisible safety restrictions allow for more precise targeting of specific objectives, enabling us to deploy quickly with very low false-positive rates."

AI researchers are that precisely targeted group.

Now forced to switch to "visible," it's purely because they got caught. They even preemptively set expectations: making it visible will "inevitably lead to more false positives." Meaning, the experience of ordinary users will have to take the hit.

This rule set was never neutral; it only protects the paymasters.

The Trifecta: Hype, Monetize, Harvest

Anthropic's playbook is more meticulously calculated than their large models themselves.

On June 10th, they first released a safety research paper. They trained a model that could reverse-engineer exploit code for vulnerabilities in a matter of hours, based on security patches. What used to take hackers days or even weeks to weaponize an N-day vulnerability is now compressed to an hour scale. The research itself is solid, but releasing it on the same day as Fable 5's launch changes the flavor: proving AI is very unsafe on one hand, while selling the "safety net solution" on the other.

The "legendary model" Fable 5 is priced at $10 per million input tokens / $50 per million output tokens, a notch pricier than Opus 4.8, with the safety classifier becoming the core premium point. Capital markets played along perfectly. Anthropic's valuation hit $96.5 billion, with plans for an October IPO underwritten by Goldman Sachs and J.P. Morgan. What they're buying isn't model parameters; it's the persona of the "safest AI company."

Research amplifies anxiety, the product harvests the premium, capital cashes out. Three moves flowing with the interests, forming a seamless loop. The only problem was, this time the loop sprung a leak: In their haste to restrict competitors, they forgot the community has people who can test for it.

OpenAI Sells Tools, Anthropic Sells Anxiety

Compared to OpenAI, the approach is completely different.

OpenAI is secretly filing for an IPO, valuation nearing a trillion, pitching the "super app": ChatGPT with 900 million weekly active users, integrating with Visa to build an ecosystem. The logic is straightforward: provide tools, earn traffic. Greedy, but candid.

Anthropic doesn't compete on scale; it competes on irreplaceability. While the whole industry is anxious about safety, it plays the role of the "only responsible adult." Its patrons are governments and giants—these are the ones most afraid of incidents and most willing to throw money at "incident prevention."

Therefore, Anthropic must keep AI perpetually in a Schrödinger's cat state of "dangerous but controllable." Too safe, and the classifier doesn't sell; too dangerous, and clients run scared. The best solution? Keep the power to define "danger" firmly in their own hands.

The dumbing-down incident just exposed this logic taken too far: the boundary of "danger" was pushed to "using Claude for AI R&D." It doesn't matter if your research is harmful; threatening their lead is the original sin.

AI has no values; it's just the boss's business spreadsheet written in code.

Apology, Just After-Sales Service for the Business

What about after the apology? Changing from secretly dumbing down to giving a heads-up before dumbing down.

Netizens see right through it: "Do you really believe it won't secretly lower output quality in the future?"

Trust, once broken, stays broken. Especially when the underlying commercial motive hasn't changed: research still amplifies anxiety, the product still harvests the premium.

The Wall Street Journal reported that OpenAI is considering significant price cuts to snatch clients from Anthropic. Price wars aren't new, but this exposes a hidden truth: The ones being downgraded covertly are AI researchers, damaging reputation among the geek community. B2B clients buying Anthropic aren't buying parameters; they're buying the persona of "the industry's safety expert." Once that persona cracks within the core developer community, why should those government and enterprise clients, who sign contracts paying a "safety premium," continue to believe you're "the safest one"?

Out of that $96.5 billion valuation, how much is solid capability, and how much is performance?

Anthropic's code is honest. The safety classifier always protects the home turf; research is responsible for amplifying anxiety; the product is responsible for harvesting the premium; the IPO is responsible for cashing out. This apology is merely a patch to the system: changing "secretly dumbing down" to "overtly dumbing down."

If safety policies really worked, Anthropic wouldn't need to publish papers every year proving patches can be breached. If the classifier were truly neutral, doing AI R&D wouldn't be classified as high-risk.

The answer was already written in the business logic.

Safety is the best business. Apology is just the after-sales service.

This article is from the WeChat public account "AI Contrarian", author: Changqing

Domande pertinenti

QWhat was the main issue with Anthropic's 'intelligent safety classifier' in the Claude Fable 5 model, according to the article?

AThe main issue was that the safety classifier would silently and automatically downgrade user requests to a weaker model (Opus 4.8) if it detected the user was conducting cutting-edge AI development or research. The article argues this was not truly about user safety but was a form of 'competitive defense' to protect Anthropic's own business advantage.

QHow does the article contrast the business strategies of Anthropic and OpenAI?

AThe article contrasts them by stating OpenAI's strategy is to 'sell tools'—focusing on building a super-app ecosystem (like ChatGPT) and monetizing scale and traffic. Anthropic's strategy is described as 'selling anxiety'—leveraging and amplifying safety concerns to position itself as the indispensable, 'most responsible' AI company for government and enterprise clients, thereby justifying premium pricing.

QWhat three-step business 'playbook' does the article attribute to Anthropic?

AThe article describes Anthropic's playbook as a three-step cycle: 1) Research that amplifies AI safety anxieties (like a paper showing models can quickly weaponize security patches). 2) Product development that harvests a price premium based on claimed safety superiority. 3) Capitalizing on this through high valuation and IPO, creating a closed financial loop.

QWhat does the article suggest is the real consequence of Anthropic's 'silent downgrade' being exposed?

AThe article suggests the real consequence is the erosion of trust, especially within the core developer and AI research community. This damage to its reputation as 'the most safety-conscious company' among technical users could ultimately undermine the 'safety premium' justification for its enterprise and government clients, threatening its business model and high valuation.

QWhat is the article's ultimate conclusion about Anthropic's concept of 'safety'?

AThe article concludes that for Anthropic, 'safety' is primarily a business strategy rather than a neutral, ethical stance. It argues that Anthropic's safety measures, such as the classifier, are designed to serve its commercial interests (like protecting its competitive lead), and that the apology was merely 'after-sales service' for this business, not a change in its underlying commercial logic.

Letture associate

The Revelation from the Raydium Theft Incident: New DeFi Vulnerabilities Lurking in Forgotten Old Contracts

**Raydium Exploit Reveals DeFi's Hidden Risk: Forgotten "Zombie" Contracts** A recent attack on Raydium's deprecated V3 AMM pools resulted in a loss of approximately $1.34 million. The hacker exploited pools that were no longer supported by Raydium's current UI or SDK but remained fully functional and accessible on-chain. This incident highlights a critical, often overlooked category of risk in DeFi: inactive or legacy smart contracts that projects fail to properly decommission. Since March 2025, there have been at least 8 publicly reported attacks targeting such abandoned contracts, with total losses around $10.8 million. Including older pools and deprecated features, the count rises to 10 incidents with roughly $22.5 million in losses. These "zombie contracts" represent a lifecycle management failure rather than a code vulnerability, yet they are typically misclassified under general "code bug" categories in security reports, masking the true scale of the problem. The root cause is that projects often merely document a contract as "deprecated" without taking essential technical steps to secure it: withdrawing remaining assets, disabling external call functions, and implementing ongoing monitoring. These forgotten, under-monitored components become prime targets for attackers. To address this, the industry needs to recognize "zombie contracts" as a distinct risk category and establish standardized decommissioning protocols. Essential steps should include: 1) a formal retirement announcement, 2) removal of all front-end integrations, 3) withdrawal of locked assets, 4) disabling key contract functions, 5) ongoing security monitoring, 6) clear user communication, and 7) a post-mortem analysis. The value of a DeFi project lies not only in its current TVL but also in the security of its historical codebase, which has now become a new attack surface.

Foresight News1 h fa

The Revelation from the Raydium Theft Incident: New DeFi Vulnerabilities Lurking in Forgotten Old Contracts

Foresight News1 h fa

Robots Begin to 'Consume Data': The Hidden Production Chain from Indian Data Factories to Billion-Dollar Humanoid Robots

Robots have started to 'consume data,' driving the formation of a new industrial supply chain focused on producing training data for embodied AI. Unlike large language models, which are trained on vast internet text corpora, embodied AI models face a 'data desert' in the physical world. This has created a massive demand for first-person perspective video data (Ego Data), captured by workers wearing cameras in places like Indian garment factories. Companies like Neocambrian AI are establishing 'data factories' where workers perform standardized tasks (e.g., sorting clothes, kitchen organization) to generate thousands of hours of video. Research, such as NVIDIA's EgoScale, demonstrates that scaling this human demonstration data predictably improves robot performance, particularly for dexterous manipulation. This has validated a training path combining large-scale human data for pre-training with smaller amounts of robot-specific data for fine-tuning. The value of different data types varies significantly, forming a 'data pyramid.' The base consists of low-cost, large-scale internet and Ego Data. Higher layers include more expensive motion-capture data (e.g., from data gloves), simulation/synthetic data, and the most costly and scarce layer: real robot teleoperation data. This demand has spawned a layered ecosystem of data suppliers: low-cost data factories, motion capture and alignment specialists, robot-native teleoperation service providers, simulation data companies, and platforms aiming for data standardization. Robot companies themselves are adopting a 'layered procurement' strategy: outsourcing generic Ego Data while building in-house capabilities for robot-specific adaptation data and the critical deployment/failure data generated in real-world applications. The industry is shifting focus from hardware and basic mobility to the data pipelines required for general-purpose capability. While parallels exist to data labeling companies like Scale AI in the LLM boom, the physical complexity of robot data—involving action success ambiguity and sim-to-real gaps—requires more integrated solutions for data collection, annotation, and a continuous feedback loop. The race is on to build the data engines that will teach robots to operate reliably in the unstructured real world.

marsbit4 h fa

Robots Begin to 'Consume Data': The Hidden Production Chain from Indian Data Factories to Billion-Dollar Humanoid Robots

marsbit4 h fa

Spicy Commentary | Michael Saylor's 'Player Talk'; 60-Year-Old Aunt Liquidated After 'Scamming a Young Man'

**"Spicy Commentary": Three Tales of Crypto's Wild Week** This week's "Spicy Commentary" column highlights three dramatic stories from the cryptocurrency world. First, **MicroStrategy's Michael Saylor** addressed the controversy over his company potentially selling Bitcoin. At the BTC Prague event, he clarified, "I never said the company can't sell Bitcoin. I told *you* never to sell *your* Bitcoin." This "do as I say, not as I do" stance was criticized by netizens as peak linguistic gymnastics, noting a history of him previously stating the company would "never" sell. Second, a **bizarre fraud case** emerged from Beijing. A 60-year-old woman, obsessed with getting rich from crypto but unwilling to risk her own savings, posed online as the 20-something "god-daughter" of a high-ranking official. She catfished a young man, convincing him to give her over 200,000 yuan for fabricated emergencies. She then invested all the stolen money into cryptocurrency with 10x leverage, only to lose everything in a market crash. The woman was sentenced to four years in prison for fraud. Finally, a **sobering trader's tale** surfaced on Reddit. A user posted "Tale of a crypto trader," confessing their net worth had plummeted from a peak of $45 million to roughly $17,200, primarily due to holding meme coins too long. The post, described as a crypto "book of confessions," sparked reactions ranging from sympathy to critique about greed, poor risk management, and the perils of treating meme coins as long-term investments instead of taking profits. The column concludes that this week featured masterful rhetoric, elaborate scams, and extreme financial volatility, stitching together another chapter in crypto's unpredictable theater.

Foresight News4 h fa

Spicy Commentary | Michael Saylor's 'Player Talk'; 60-Year-Old Aunt Liquidated After 'Scamming a Young Man'

Foresight News4 h fa

Tremble Humans, AI Continues Its Accelerated Sprint

Trembling, Humans: AI Continues Its Accelerated Sprint Yes, AI is still rapidly accelerating. While deep learning seemed to stall quickly in its early years, large models after years of development show no sign of hitting their ceiling. At the Zhiyuan Conference 2026, the focus is on enabling AI to move from the digital world into the physical world. Scaling Law remains effective, continuing to drive advancements in both large language models and multimodal models. The industry is now entering a phase of pursuing World Models, though unresolved technical paths and data issues mean this exploration may take 3-5 more years. Concurrently, breakthroughs in Agents are accelerating AI's real-world application in fields like healthcare and meetings. Making Agents truly useful requires key hardware-software co-design, evident from the strong presence of chip vendors at the conference. We stand at a new historical threshold where AI is becoming a foundational force reshaping the world. The first day of the conference highlighted AI's evolution from "knowing how to chat" to "knowing how to work." Scaling Law persists, World Models are the next key battleground, and Agents are transitioning from usable to好用 (user-friendly). Scaling Law is not ending but diversifying. New models like Anthropic's Fable 5 demonstrate scaling through parameter size, synthetic data, and reinforcement learning. Advancements in AI Coding and Agent deployment are enabling a trend of AI self-evolution, potentially allowing AI to take over digital world iterations. World Models represent the next frontier for large models extending into the physical realm, but no current model is truly impressive at solving real-world problems. Technical consensus is lacking, with debates on data sources (video, simulation, real-world). Different approaches are emerging: language-centric, pixel-centric, 3D-structure-centric, and visual-representation-centric models. Zhiyuan Institute is exploring a fifth path: unified latent space modeling fusing language and visual representations, and introduced its own under-development World Model, Physis-v0.1. On the product side, Agents are key to bringing AI into daily life. Since 2025, the "Year of the Agent," products have become more proactive and capable of complex tasks. Zhiyuan showcased four vertical Agents for cardiac diagnosis, autonomous research, meeting summarization, and protein risk discovery. However, technical challenges remain, particularly in context engineering like memory and orchestration. "Harness" – the engineering framework around an Agent – is crucial for maximizing its capabilities by clarifying intent, designing workflows, and incorporating validation and feedback. In summary, AI's breakneck pace continues on multiple fronts: foundational model scaling, the ambitious pursuit of World Models for physical understanding, and the ongoing refinement of practical Agents. The journey from capable to truly reliable and useful AI systems is well underway.

marsbit4 h fa

Tremble Humans, AI Continues Its Accelerated Sprint

marsbit4 h fa

Trading

Spot
Futures
活动图片