Google DeepMind describes AlphaProof Code, combining human-approved specifications with agent-generated proofs checked by Lean. Its libsignal study proves specified properties of PQXDH and the Triple Ratchet at the AArch64 machine-code level. Assurance depends on the specifications, formal semantics and other stated trust assumptions; it does not cover every property of the Signal application.
Google DeepMind
Google's AI research lab.
A Google DeepMind preprint tests collaborative attacker agents on a synthetic instruction-hierarchy benchmark with randomly selected, non-malicious tool-call goals. Stronger attackers and larger query budgets expose failures that static tests miss. The experiments isolate model behavior and do not measure successful real-world data theft or a complete deployed defense stack.
Google’s October 28 announcement reports unguarded Gemini 4 Carbon cyber evaluation scores and phased defender access through Fairwind. Its November 10 release expands access to paying developers and enterprises and rolls out consumer access to Google AI Ultra subscribers. Broader general availability does not imply identical safeguards or access to the specialized cyber configuration.
Google Project Zero compares BigSleep and Fuzzilli on V8 over April–September 2026, reporting 45 versus 37 actionable vulnerabilities and about 25% observed overlap. A separate historical rediscovery experiment finds higher overlap; the report does not establish a controlled cost-efficiency ranking.
OpenAI describes coordinating fixes for two critical PHP flaws after Calif identified one using AI and Wiz independently found both; PHP released fixes on October 21, and OpenAI proposed a disclosure framework. Follow-up evidence includes CISA’s October 26 listing of both flaws as known exploited and Palo Alto Networks’ report of active exploitation against internet-exposed management interfaces. Wiz’s October 27 report describes declining observed exposure and failed chains in its own captured attempts; this does not establish an absence of exploitation elsewhere or a global compromise count.
Google DeepMind announced Gemini 4 Argon on 2026-09-30 and says it is rolling out to a set of trusted cyber defenders through the Fairwind Program, and says it will release the model without cyber guardrails for those defenders and Google's internal teams. Google claims the model can autonomously find, validate and patch critical vulnerabilities, ties for first on CWE-bench v1 at 68%, and improves on Gemini 3.8 Flash Cyber in vulnerability discovery, without giving numbers for the comparison. It says broad release awaits further safeguard work in four areas (misuse, prompt injection, misalignment and hardening), citing its Frontier Safety Framework for the misuse safeguards; all of these are Google-reported claims that have not been independently verified.
Google introduced Gemini 3.8 Flash Cyber, a cybersecurity-tuned model with more permissive cyber mitigations, available only to trusted defenders through a new Fairwind Program. Google says it prioritized vulnerability fixing over exploitation and reports 47.2% pass@1 on Collinear's CWE-Bench patching benchmark, over 70% on an internal 20-language discovery benchmark, and 2.6 times more correct Chrome patches than larger commercial models.
UK AISI's new Control Red Team tested an asynchronous reasoning monitor with Google DeepMind and successive versions of Anthropic's synchronous internal agentic-coding monitor, and reports vulnerabilities in each. Successful attacks exploited gaps in monitor prompts, weaknesses in the underlying LLM, and hard-coded rules that decide which actions reach the monitor. In a separate early experiment, an automated evolutionary search against a Claude 4.6 Sonnet monitor found attack transcripts scoring 3/100 on suspicion, versus 14-18 for best-of-N baselines. That experiment used a single-step coding task in which the agent covertly exposes its API key.
Google DeepMind introduces an AI Control Roadmap for internally deployed agents. It treats potentially misaligned AI as an insider threat, builds a threat model on MITRE ATT&CK tactics and techniques, and scales detection and response requirements with levels of agent evasion (D1-D4) and attack capability (R1-R3). It reports analyzing about a million coding agent tasks with a prototype asynchronous monitor, which informed a live monitor for its Gemini Spark agent. It says the majority of flagged events were not adversarial and often stemmed from misinterpretation or overeagerness.
Google DeepMind's Frontier Safety Framework version 3.1 introduced Tracked Capability Levels for earlier warning in CBRN and ML R&D and misalignment, and raised the recommended security for the CBRN, cyber and harmful manipulation CCLs to Security Level 2+. The cyber CCL definition itself, Cyber uplift level 1, is unchanged from v3.0.
Google DeepMind introduced CodeMender, an agent built on Gemini Deep Think models that combines static and dynamic analysis, fuzzing, differential testing and SMT solvers with LLM-based critique to generate and validate security patches. DeepMind reports 72 security fixes upstreamed to open-source projects over six months, all reviewed by human researchers before submission; in May 2026 Google said it would fold CodeMender into its enterprise agent platform.
Google DeepMind's Frontier Safety Framework version 3.0 adds a harmful manipulation CCL and expands misalignment and internal-deployment provisions. Its cyber domain keeps a single CCL, Cyber uplift level 1, for models providing sufficient uplift with high-impact cyber attacks to add expected harm at severe scale, paired with Security Level 2; the framework reasons that automated cyber-defense and social adaptation make higher security levels likely unwarranted.
Google's vice president of security announced that Big Sleep had reported 20 vulnerabilities, mostly in open-source projects such as FFmpeg and ImageMagick, with details withheld pending fixes. A Google spokesperson told TechCrunch each flaw was found and reproduced by the agent without human intervention, with a human expert reviewing reports before submission.
Google reports that, working from Google Threat Intelligence information, the Big Sleep agent found a critical SQLite memory-corruption flaw (CVE-2025-6965) that Google says was known only to threat actors and at risk of exploitation. Google says it reported the flaw for patching before attackers could exploit it, says it believes this is the first time an AI agent directly foiled an in-the-wild exploitation effort, and says Big Sleep is being applied to open-source projects.
Shi and colleagues describe Google DeepMind's continuous adaptive-attack evaluation of Gemini against indirect prompt injection in tool-use settings. On Gemini 2.0, adaptive attacks generally matched or beat non-adaptive ones against eight baseline defenses, reaching 98.4% against in-context learning and 82.4% against spotlighting, while a warning defense and a user-instruction classifier held (at most 10.8% and 3.0%). Adversarial fine-tuning for Gemini 2.5 lowered but did not eliminate attack success.
Debenedetti and colleagues (Google, Google DeepMind, ETH Zurich) propose CaMeL, which extracts control flow from the trusted user query so untrusted data cannot change which actions run, and attaches capabilities to data to block unauthorized flows. On AgentDojo the first version reported 67% of tasks solved with provable security; the June 2025 revision, with newer models, reports 77% versus 84% for an undefended system.
Google Project Zero and Google DeepMind report that Big Sleep, an LLM agent evolved from Project Naptime, found an exploitable stack buffer underflow in SQLite during a variant-analysis experiment seeded with newly landed commits. The bug was reported in early October 2024 and fixed the same day, before it appeared in an official release; Google believes this was the first public example of an AI agent finding such a memory-safety flaw in widely used software.