According to @AnthropicAI, internal classifiers reduced Claude jailbreak success from 86% to 4.4%, indicating a substantial decrease in successful exploits. Source: @AnthropicAI on X, Jan 9, 2026, https://twitter.com/AnthropicAI/status/2009739654833029304 According to @AnthropicAI, the classifiers were expensive to run, impacting operational cost profiles for deployments. Source: @AnthropicAI on X, Jan 9, 2026, https://twitter.com/AnthropicAI/status/2009739654833029304 According to @AnthropicAI, the system became more likely to refuse benign requests after adding the classifiers. Source: @AnthropicAI on X, Jan 9, 2026, https://twitter.com/AnthropicAI/status/2009739654833029304 According to @AnthropicAI, despite improvements, the system remained vulnerable to two types of attacks shown in their accompanying figure. Source: @AnthropicAI on X, Jan 9, 2026, https://twitter.com/AnthropicAI/status/2009739654833029304

According to VitalikButerin, there are five main concerns in the crypto space that people often express worry about. Instead of feeling overwhelmed by these challenges, likening them to facing an ELO 4500 chess player, Buterin suggests that the crypto community should focus on identifying specific attack vectors and systematically addressing each one. This proactive approach could help mitigate potential risks in the cryptocurrency market.

According to Vitalik Buterin, instead of feeling overwhelmed by potential threats in the cryptocurrency space, akin to being outmatched by an extremely high-ranked chess player, it is more productive to identify specific attack vectors and address them individually. This proactive approach can help mitigate risks in the crypto market.