Anthropic's Claude Fable 5.1 appears to have cracked a centuries-old number puzzle that researchers considered unsolved.
The article Claude Fable 5.1 decoded a centuries-old royalist message hidden in plain sight since 1653 appeared first on The Decoder.
Earlier this week, researchers outlined an attack that used a secret input provided by Microsoft 365 Copilot for enterprise to cause the AI assistant to exfiltrate a password present in the user’s inbox. Now, a separate team has devised a similar attack against Grok. The new data theft hack employs a deceptively simple trick to force the Elon Musk-owned large language model to steal user chats and other personal information. At the time this post went live, the assistant continued to cough up the data, despite xAI being informed of it in June.
The lesson from both this week’s episodes—and the countless other ones that have come before it—is that LLMs are incapable of solving the root causes for prompt injections, the most severe vulnerability classes they’re most prone to. That leaves AI developers with no other option but to build a guardrail that steers the model away from the harmful actions. As I noted in Tuesday’s story, the approach is tantamount to a road traffic safety engineer erecting a protective rail around a dangerous bend rather than banking the curve.
Prompt injections exploit LLMs' training to comply with user requests whenever possible. Attackers can capitalize on the predilection by smuggling harmful instructions into emails or webpages the assistant is instructed to summarize. Because LLMs can’t reliably distinguish between content in an email sent by an untrusted party and user instructions entered directly into a prompt, the overly solicitous LLM faithfully follows them. To date, Grok and other LLMs' only recourse is to create guardrails that flag suspicious instructions and forbid them from being executed.
© Getty Images | SOPA Images
The White House is drastically shortening the deadline for government agencies and organizations to adopt new quantum-resistant encryption systems that will withstand attacks that use quantum computers, as the federal government seeks to protect decades’ worth of secrets belonging to militaries, banks, governments, and most individuals on Earth.
The executive order, titled Securing the Nation against Advanced Cryptographic Attacks, requires computing systems for “high-value assets” and “high-impact systems” to transition to post-quantum cryptographic key establishment schemes by December 31, 2030, and to quantum-safe digital signature schemes by December 31, 2031.
The new deadline, which for many organizations is about five years sooner than the previous one, comes on the heels of recent research showing that the resources and cost for building a cryptographically relevant quantum computer are far less than previous consensus estimates. In response, Google, Cloudflare, and other companies recently tightened their timelines for moving off vulnerable systems to 2029.
© vital
The Texas Attorney General has sued Meta over allegations that the company’s WhatsApp messenger, used by more than 3 billion people, doesn’t provide the end-to-end encryption (E2EE) it has long claimed.
Since at least 2016, Meta (then named Facebook) has said WhatsApp provides robust end-to-end encryption, meaning that messages are encrypted on a sender’s device with keys that are available only to the receiver's. By definition, E2EE means that no one else—including the platform itself—can read the plaintext messages.
In sworn testimony before two US Senate committees in 2018, CEO Mark Zuckerberg said Meta does “not see any of the content in WhatsApp; it is fully encrypted” and that “Facebook systems do not see the content of messages being transferred over WhatsApp.” The engine for this E2EE is the Signal protocol, an open source code base that multiple third-party experts have said lives up to its promises.
© Getty Images
Login