Person
Bruce Schneier is a security technologist, cryptographer, and author who writes widely on security, privacy, and surveillance at his blog schneier.com.
Forums are hiding invisible Unicode prompt-injection traps on signup pages to make AI agents give themselves away, according to a dev.to post. Teams running browsing agents now have to defend against hostile page text from site operators, not only from attackers.
Reality
UC San Diego and Inria researchers forged a 1024-bit RSA signature through an HSM in 1,380 core-years, against at least 500,000 to factor the key. The attack only works with raw, unpadded signing access, so how exposed a custodian is depends on how its HSM is configured.
Reality
Their Lawfare essay says the architecture built after 9/11 now runs largely on data that companies gather for their own business, and it cites FBI Director Kash Patel's testimony that the bureau buys information on Americans from brokers.
Reality
A scan of 6,214 corporate domains found 120 sites whose AI-facing docs point at code packages and domains nobody owns, and registering a few of those names produced installs traced back to Claude, Codex and Hermes agents inside real networks.
Reality
A sender calling itself an autonomous Claude instance mailed Bruce Schneier a door-by-door account of trying to turn $4.75 into $10 in a day, in which captchas, datacenter IP reputation and payment settlement held while identity verification never engaged.
Reality
Freezers failed at more than a half-dozen bases over a few nights, and the test now is whether anyone can tell a coordinated shut-off from a fleet of old equipment failing together. So far the published evidence supports neither reading over the other.
Perspective Coverage
3 publishersReality
Anthropic published two AI-derived cryptanalysis results in July and OpenAI's model disproved an 80-year-old conjecture in May, but both authors argue the models are still weak exactly where deep new theory is required.
Reality
AI models generated hundreds of thousands of phage designs and 16 of the built ones replicated. Once a genome can be authored rather than copied, matching orders to known threats stops being enough.
Reality
A benchmark called CIMemories reports frontier models pushing sensitive attributes into tasks that do not need them, with GPT-5 violations climbing from 0.1% at one task to 9.6% across 40.
Reality
Schneier and Sanders say June's IPO filings ended governance by founder intent at OpenAI and Anthropic. The operator question underneath it is what happens to a dependency that cannot be priced.
Reality