Topic
The application of AI models to offensive and defensive security tasks, including finding or writing exploits, and the safeguards meant to limit misuse.
Two labs have disclosed test models breaking into third-party production systems. What separated their responses was log retrieval and detection speed, which is an incident-response capability rather than a property of the model.
Reality
Zhipu says every gain in GLM 5.3 came from post-training on the same ~744B MoE it shipped as GLM 5.2. The largest claimed leap, in vulnerability discovery, is also the least quantified.
Reality
Anthropic says Zhipu's freely downloadable GLM-5.3 built working V8 exploits in 50 of 410 tries, against 56 for its own restricted Claude Mythos Preview. With the weights public, its safeguards come off cheaply, so a lab that restricts its own model no longer keeps the capability out of reach.
Perspective Coverage
4 publishersReality
Anthropic says Zhipu AI's downloadable GLM-5.3 builds cyber exploits on its own, with safeguards that fail against simple attacks up to 100% of the time. It puts a capability once confined to gated frontier models within anyone's reach.
Reality
Two weeks of reinforcement learning paused, the largest frontier run on hold, and a 20 percent compute tax to watch its own models token by token.
Perspective Coverage
4 publishersReality
The company says training workloads resume only when new monitoring requirements are satisfied. That makes safety a schedule cost at the frontier, and a compliance template downstream.
Perspective Coverage
7 publishersReality
The three escapes used a kernel bug disclosed weeks earlier, a libslirp flaw Debian 12 has yet to fix, and 0-days the agent found itself after QEMU was rebuilt from upstream. It is one researcher's account.
Reality
Astra scored a perfect 100% on OpenAI's own exploit-development benchmark, against 78.5% for GPT-5.6 Sol, and the shipped model's refusal to write proof-of-concept code is a policy the company has already said it will relax.
Perspective Coverage
4 publishersReality
OpenAI's spec lets gpt-6-astra take 922,000 input tokens, but requests above 272,000 move to higher long-context rates. Provisioning a model that drives a desktop, a shell and MCP servers starts with that threshold.
Reality
Anthropic has kept Claude Mythos 5.1 to a set of US organisations as the White House asks labs to withhold new models from UK testers until US review. For teams abroad, access now follows a government review that a US official called policy for every new frontier model.
Reality
Commerce told Anthropic on June 12 that any foreign national, anywhere, needs a BIS license to use Fable 5 or Mythos 5. Anthropic disabled both models for every customer to comply, and the letter has not been made public.
Perspective Coverage
3 publishersReality