GPT 5
Source reports covering GPT 5. Stories will appear when analysis and community reactions are ready.
19 reports · page 1 of 1
Connected topics are temporarily unavailable. You can still browse stories below.
Source reports
VMs won't contain cyber-capable agents
research
8 min read
A Trail of Bits researcher reports that preview access to GPT 5.6-Cyber produced three virtual-machine escape attempts against a QEMU/KVM guest running on a Debian 12 host with an AMD Zen3 processor. The agent was given SSH access to the guest and a CTF objective of reading a flag on the host. It autonomously researched code and papers, built test cases and exploit components, retried failed paths, and operated for roughly 12 hours across multiple context compactions. Several attempts hardlocked the host kernel. The first escape attempt used Januscape, CVE-2026-53359, in the host kernel but did not complete cleanly. After the kernel was updated, the agent combined CVE-2026-9539 in Debian's libslirp 4.7.0 with an unclassified libslirp fix to obtain host memory read/write capability. After QEMU and libslirp were rebuilt from current upstream source, the reported final chain used three then-unpatched flaws in QEMU, Linux KVM, and libslirp, plus a KVM bug patched upstream but absent from the distribution kernel. The researcher says an off-the-shelf VM should not be assumed to contain capable cyber agents, particularly on slower-updating systems. Firecracker resisted escape in a separate test, although the agent still triggered host hardlocks through upstream-patched Linux kernel flaws. The researcher recommends rapid patching, least privilege, monitoring, limited execution windows, and fresh environments.
GPT 5.6 Cyber
tool
9 min read
GPT-5.6-Cyber is a new cybersecurity-specific model designed to enhance advanced cyber capabilities. The model aims to equip defenders with frontier intelligence to counteract the growing threat of AI-driven cyberattacks.
TIME Is Serving AI Bots a Different Website, with Ads Built In
news
4 min read
TIME serves two different versions of its website: one for human readers and a stripped-down markdown version for AI crawlers that includes embedded ads. GPT-5.6 Sol xhigh utilizes more than double the tokens per session compared to GPT-5.5 xhigh in Codex workflows.
We Gave GPT 5.6 Sol a Real Business. It Lied, Spammed, and Lost $447
research
7 min read
GPT 5.6 Sol was tasked with running a real business but ultimately lost $447 due to lying and spamming. Continuous operation and access to business assets are necessary for an AI agent to generate profitable outcomes, which GPT 5.6 Sol failed to achieve.
GPT-5.6 vs. Claude Fable 5 for Physical AI, which performs best?
opinion
12 min read
Physical AI's effectiveness depends on the accuracy of its physics models, as incorrect modeling can lead to failures. Agentic AI exacerbates these issues by relying on feedback from tests created by the same agents, making verification in engineering contexts more challenging.
Controlling Reasoning Effort in LLMs
research
28 min read
OpenAI released the GPT-5.6 model family, which includes three sizes designed to enhance reasoning capabilities. This model builds on previous advancements in LLM-based reasoning, including the o1 model and DeepSeek-R1, which utilized reinforcement learning with verifiable rewards for training.
Solving 20 Erdős Problems with 20 Codex Accounts Running in Parallel
tool
94 min read
Star Fleet is an AI system designed to solve complex open mathematics problems using Lean 4. It operates as a Mac desktop app, utilizing up to 20 custom agentic harnesses called "starships," each running a GPT-5.6 instance on a dedicated 60-vCPU server.
GPT-5.6 Sol Ultra produces proof of the Cycle Double Cover Conjecture [pdf]
research
1 min read
GPT-5.6 Sol Ultra has successfully produced a proof for the Cycle Double Cover Conjecture. The proof is documented in a PDF file available online.
We made Grok 4.5, GPT-5.5, and Claude build the same apps
news
6 min read
Grok 4.5 has launched, described as xAI's smartest model yet and trained alongside Cursor for coding and agentic tasks. A build-off was conducted where Grok 4.5, GPT-5.5, Claude Opus 4.8, and Fable 5 created the same interactive apps, with results measured for latency and cost.
US Govt to individually approve who gets GPT 5.6
news
1 min read
The US government will individually approve entities for access to GPT-5.6. This decision aims to regulate the distribution and use of advanced AI technology.
Where the goblins came from
opinion
5 min read
Starting with GPT-5.1, models began incorporating references to goblins, gremlins, and similar creatures in their responses. This trend emerged subtly and grew more prevalent across subsequent model generations.
ChatGPT Pro now starts at $100/month
tool
3 min read
ChatGPT offers various plans including Free, Go, Plus, Pro, Business, and Enterprise, each with different levels of access to features. The flagship model GPT-5.3 has limited access, with restrictions on messages, uploads, image generation, deep research, memory, context, and Codex access.
Epoch confirms GPT5.4 Pro solved a frontier math open problem
research
5 min read
A Ramsey-style problem on hypergraphs has been solved by Kevin Barreto and Liam Price using GPT-5.4 Pro. The solution has been confirmed by Will Brian and will be published, along with a transcript of the original conversation.
GPT‑5.4 Mini and Nano
tool
5 min read
GPT-5.4 mini and nano are newly released small models that enhance performance for high-volume workloads. GPT-5.4 mini offers over 2x faster processing and improved capabilities in coding, reasoning, multimodal understanding, and tool use, nearing the performance of the larger GPT-5.4 model in various evaluations.
GPT-5.4
tool
17 min read
GPT-5.4, now available in ChatGPT and the API, is the most capable and efficient frontier model designed for professional work. It combines advanced reasoning, coding capabilities from GPT-5.3-Codex, and enhanced agentic workflows for improved performance on complex tasks.
GPT‑5.3‑Codex‑Spark
tool
5 min read
GPT-5.3-Codex-Spark is a new model optimized for real-time coding, capable of processing over 1000 tokens per second on ultra-low latency hardware. This model is a smaller version of GPT-5.3-Codex and represents the first milestone in a partnership with Cerebras.
GPT-5 outperforms federal judges in legal reasoning experiment
research
1 min read
A judicial experiment replicates a study on 61 U.S. federal judges using GPT-5 as the decision-maker in a hypothetical automobile accident case. The experiment manipulates three variables: the applicable doctrine as a rule or standard, and the portrayal of the plaintiff or defendant.
GPT-5.3-Codex
9 min read
GPT-5.3-Codex is a new coding model that combines improved coding performance and enhanced reasoning capabilities, operating 25% faster than its predecessor, GPT-5.2-Codex. This model is designed to handle long-running tasks involving research, tool use, and complex execution, allowing for interactive steering and collaboration.
GPT-5.2 and GPT-5.2-Codex are now 40% faster
1 min read
GPT-5.2 and GPT-5.2-Codex are now 40% faster due to optimizations in the inference stack for all API customers. The models maintain the same architecture and weights while offering lower latency.