METR称Anthropic AI研发提速或于7月触及预警门槛
METR调查称，Anthropic模型能力进步速度约为此前两倍，相关研发提速门槛可能早在7月触及。多数参与评估者仍认为当时尚未越线。
Loading…
bibibi.ccbibibi.ccLoading…
Connect scattered updates into a story you can understand.
METR调查称，Anthropic模型能力进步速度约为此前两倍，相关研发提速门槛可能早在7月触及。多数参与评估者仍认为当时尚未越线。
美国司法部宣布成立超级智能特别工作组，由两名官员牵头，防范外国行为者、犯罪组织及欺诈者滥用超级智能技术。
《纽约时报》采访20多名知情人士，报道Dario Amodei的健康经历、慈善观念、AI安全立场及其离开OpenAI的内部冲突。
报道称，OpenAI总裁Greg Brockman已停止向政治组织LTF追加捐款，承诺的5000万美元中实际支付2500万美元。该组织的募资与部分竞争性选举表现亦低于早期预期。
赵长鹏在牛津辩论社访谈中称，近期黑客事件本可防范；AI短期会暴露更多漏洞，但长期有助于提升系统安全。
据IT之家报道，Anthropic正开发开源系统FlashProve，以密码学方式验证AI输出所来自的模型，并支持本地验证。
PANews援引X安全团队官方账号@Safety称，X已任命Michael O'Herlihy为新任安全主管，负责平台及AI系统的用户和品牌安全工作。
报道称，Anthropic在11月2日更新的IPO招股书中称，未来AI或跨过递归自我改进门槛并加快研发；公司称现有模型尚未达到该水平。
South Korea’s Financial Services Commission reportedly will November 5 convene a financial-sector AI security threat response roundtable, attended by the heads of the country’s five major banks.
Anthropic disclosed that, since this year, Claude June has reviewed abuse risks in conversations in the background after responding, but the mechanism is currently used primarily for post hoc monitoring rather than real-time blocking.
PANews relayed a a16z survey of 100 participants CISO which stated that 87%enterprises expect annual 2027 growth in cybersecurity budgets, with the defensive focus shifting to automated closed-loop remediation.
PANews said that Anthropic stated on X that it has trained Claude to assess potential misuse in conversations after completing a response.
